Biclustering in bioinformatics using big data and High Performance Computing applications: challenges and perspectives, a review

Vol.:(0123456789) The Jou nal o Supe compu ing (2025) 81:1123 h ps://doi.o g/10.1007/s11227-025-07563-6 Biclus e ing inbioin o ma ics using big da a andHigh Pe o mance Compu ing applica ions: challenges andpe spec i es, a e iew Au elioLópez‑Fe nández4· F anciscoA.Gomez‑Vela1· DomingoS.Rod iguez‑Baena1· Fe nandoM.Delgado‑Cha es2· Jo geGonzalez‑Dominguez3 Accep ed: 6 June 2025 © The Au ho (s) 2025 Abs ac Biclus e ing is a powe ul machine lea ning echnique ha simul aneously g oups ows and columns in ma ix-based da ase s. Applied o gene exp ession da a in bio- in o ma ics, i s use has expanded alongside he apid g ow h o high- h oughpu sequencing echnologies, leading o massi e and complex biological da ase s. This e iew aims o examine how biclus e ing me hods and hei alida ion s a egies a e e ol ing o mee he demands o High Pe o mance Compu ing (HPC) and Big Da a en i onmen s. We p esen a s uc u ed classi ica ion o exis ing app oaches based on he compu a ional pa adigms hey employ, including MPI/OpenMP, Apache Hadoop/Spa k, and GPU/CUDA. By syn hesising hese de elopmen s, we highligh cu en ends and ou line key esea ch challenges. The knowledge ga he ed in his wo k may suppo esea che s in adap ing and scaling biclus e ing algo i hms o analyse la ge-scale biomedical da a mo e e icien ly. Ou con ibu ion is in ended o b idge he gap be ween algo i hmic inno a ion and compu a ional scalabili y in he con ex o bioin o ma ics and da a-in ensi e applica ions. Keywo ds Biclus e ing· Big da a· High Pe o mance Compu ing· Bioin o ma ics Abb e ia ions ACV A e age co ela ion alue AWS Amazon web se ices CC Cheng–Chu ch CCS The condi ion-dependen co ela ion subg oups CPU Cen al p ocessing uni CUDA Compu e uni ied de ice a chi ec u e EBI Eu opean bioin o ma ics ins i u e EBIC E olu iona y-based bIClus e ing ENCODE Encyclopedia o DNA elemen s Ex ended au ho in o ma ion a ailable on he las page o he a icle A.López-Fe nández e al. 1123 Page 2 o 52 FCA Fo mal concep analysis FLOC FLexible o e lapped biClus e ing FPGA Field-p og ammable ga e a ay GB Gigaby e GBC Geome ic biclus e ing algo i hm GMQL GenoMe ic que y language GPGPU Gene al-pu pose compu ing on g aphics p ocessing uni s GPU G aphics p ocessing uni GTEX Geno ype- issue exp ession p ojec HDFS Hadoop dis ibu ed ile sys em HPC High Pe o mance Compu ing KEGG Kyo o encyclopedia o genes and genomes LCS Longes common sequence MATLAB MAT ix labo a o y MCC Ma hews co ela ion coe icien MPI Message passing in e ace MR MapReduce MSR Mean squa e esidue NGS Nex -gene a ion sequencing NMF Non-nega i e ma ix ac o isa ion NMI No malised mu ual in o ma ion PBD-SPEA2 Pa allel biclus e ing de ec ion PC Pa allel coo dina e PCR Polyme ase chain eac ion PGAS Pa i ioned global add ess space POSIX Po able ope a ing sys em in e ace PPI P o ein–p o ein in e ac ion RAM Random access memo y RDD Robus dis ibu ed da ase s RNA Ribonucleic acid SM NVIDIA S eaming mul i-p ocesso SMP Sha ed memo y pa allelism SPEA2 S eng h pa e o on e olu iona y algo i hm2 SPMD Single p og am, mul iple da a SPUs S eam p ocesso uni s SPs S eaming p ocesso s SQL S uc u ed que y language STRING Sea ch ool o he e ie al o in e ac ing genes/p o ein TCGA The cance genome a las UM Uni ied memo y Biclus e ing inbioin o ma ics using big da a andHigh… Page 3 o 52 1123 1 In oduc ion Biclus e ing echniques ha e been success ully applied o se e al esea ch a eas, including ene gy consump ion [1], economics [2], ading o ecas ing [3], ma ke - ing [4], o ecommenda ion sys ems [5]. In bioin o ma ics, hese echniques a e o en used o applica ions such as analysing gene exp ession da a, disco e ing and anno a ing new unc ionali ies o unclassi ied genes, inding exp ession modules, econs uc ing biological ne wo ks, elucida ing disease mechanisms, and s a i ying pa ien s [6, 7]. The applica ion o biclus e ing echniques on gene exp ession da a- se s has wo main ad an ages o e clus e ing: (a) i g oups bo h genes and expe i- men al condi ions, which is much close o biology since a subse o genes may ha e common biological beha iou only unde a subse o expe imen al condi ions o samples [8], and (b) i conside s g oup o e lapping, allowing genes o con ibu e o mo e han one biological ac i i y. Despi e hese ad an ages, se e al challenges a ise, mainly due o he exponen ial g ow h o biological da ase s—in bo h olume and complexi y—gene a ed by nex -gene a ion sequencing (NGS) echnologies [9] and he c ea ion o la ge-scale genomic conso iums [10]. The comple e anno a ion and quan i ica ion o he exp ession le el o all genes and hei iso o ms in a speci ic sample [11], along wi h he da a olume ha ep- esen s gene sequencing in o ma ion om 100 million o 2 billion indi idual pa ien s in he yea 2025 [10], demons a e some challenges aced in his ield. Also, i is essen ial o conside ha his da a may o igina e om a ious sou ces, be s o ed in se e al o ma s, and ypically con ain noise and a e ac s ha equi e esolu ion [12]. This inc eases he compu a ional complexi y o biclus e ing me hods used o e alua e and ex ac meaning ul in o ma ion om la ge biomedi- cal and biological da ase s. To add ess hese challenges, High Pe o mance Compu ing (HPC) app oaches, such as Gene al-Pu pose Compu ing on G aphics P ocessing Uni s (GPGPU), Big Da a amewo ks, and adi ional pa allel and dis ibu ed compu ing pa a- digms, ha e been adop ed o imp o e pe o mance and scalabili y [13]. Despi e ha ing p o en highly use ul in mul iple con ex s, con en ional biclus e ing ech- niques exhibi a numbe o echnical and compu a ional limi a ions when dealing wi h ex emely la ge da ase s. These limi a ions can be summa ised as ollows: • High compu a ional complexi y: Con en ional echniques o en aim o ind op imal solu ions a he cos o high compu a ional complexi y. This becomes unsus ainable when inpu ma ices con ain millions o ows and columns. Fo example, he i s ue biclus e ing algo i hm, he Cheng–Chu ch algo i hm [14], has app oxima ely quad a ic complexi y in he numbe o ows and col- umns, making i poo ly scalable. • Memo y limi a ions: Many algo i hms s o e en i e ma ices o mul iple auxil- ia y s uc u es in memo y, which becomes un easible wi h e y la ge da ase s. Fo ins ance, algo i hms like SAMBA [15] may encoun e memo y manage- men issues when handling gene exp ession da ase s wi h mo e han 100,000 genes and condi ions. A.López-Fe nández e al. 1123 Page 4 o 52 • Sensi i i y o noise and edundancy: In la ge da ase s, noise is ine i able and can lead o many spu ious pa e ns. Classical echniques a e no always designed o e ec i ely il e ou his noise. Algo i hms such as Bimax [16] sea ch o exac biclus e s (i.e. noise- ee), which makes hem less use ul o la ge-scale eal-wo ld da a. In addi ion o hese h ee main issues, he e is also a lack o adap abili y o dis- ibu ed s uc u es and an inabili y o handle he e ogeneous o dynamic da a. The impo ance o s udying biclus e ing unde he Big Da a and HPC pa a- digms lies in he g owing need o ex ac biologically meaning ul pa e ns om massi e and complex da ase s. Fo example, in he wo k p esen ed by Liu e al. [17], i e gene exp ession da ase s we e p ocessed, gene a ing a o al o 8280536 biclus e s. F om a genomic conso ium poin o iew, ano he example can be ound in The Cance Genome A las (TCGA) [18], which p o ides mul i-omics da a o housands o umou samples ac oss nume ous cance ypes. The iden- i ica ion o biclus e s in his con ex can help disco e cance -speci ic gene exp ession modules, s a i y pa ien s based on molecula p o iles, and suppo he design o a ge ed he apies. Howe e , he olume and complexi y o hese ypes o da ase means ha adi ional biclus e ing me hods a e insu icien . So a sys- ema ic examina ion o new biclus e ing app oaches ha in eg a e obus and e i- cien compu a ional s a egies wi h biological in e p e abili y is equi ed. Since he publica ion o he wo k by Sa a C. Madei a e al. [19], many s ud- ies ha e been p esen ed ecen ly whose objec i e is o p esen a classi ica ion o compa a i e s udy. Despi e his weal h o in o ma ion, o he bes o ou knowl- edge, none o hese pape s has ocused hei s udy on how new o e en exis ing biclus e ing echniques ha e me he challenge o being able o adap o a Big Da a o HPC ecosys em o p ocess he biological and biomedical da ase s ha a e cu en ly a ailable. The e is also no ela ed wo k on he abili y o biclus e ing echniques o alida e he la ge numbe o esul s hey gene a e o how biclus e alida ion me hods add ess his p oblem. Mo eo e , in o de o cope wi h a Big Da a ecosys em, hese echniques should also ake in o accoun he a ious chal- lenges hey mus ace when ying o adap o some speci ic HPC me hods. How- e e , he e a e also no wo ks ha add ess he challenges in ol ed in his ype o adap a ion. This pape p esen s an o e iew o esea che s and p ac i ione s o disco e sui - able biclus e ing solu ions o la ge-scale bioin o ma ics applica ions while p omo - ing u u e esea ch ha connec s biological knowledge wi h compu a ional scalabil- i y. This wo k explo es he e olu ion o biclus e ing echniques in esponse o he inc easing scale and complexi y o bioin o ma ic da ase s. I examines he inco po- a ion o High Pe o mance Compu ing (HPC) and Big Da a amewo ks as key ena- ble s o imp o ing he scalabili y and applicabili y o hese me hods in eal-wo ld biomedical scena ios. Fu he mo e, i analyses he p incipal limi a ions and ongoing challenges ha pe sis in his in e disciplina y ield, highligh ing he need o con in- ued me hodological inno a ion o ensu e bo h compu a ional e iciency and biologi- cal in e p e abili y. The e o e, we can summa ise he con ibu ions o his wo k as ollows: Biclus e ing inbioin o ma ics using big da a andHigh… Page 5 o 52 1123 • A sys ema ic and up- o-da e su ey o biclus e ing echniques ha add ess he challenges posed by la ge-scale biomedical da a. • A classi ica ion o hese me hods based on hei compu a ional pa adigms and p ac ical implemen a ions. • A c i ical analysis o hei biological applica ions and he epo ed compu a ional pe o mance based on a comp ehensi e e iew o he exis ing li e a u e. • A summa y o cu en challenges and u u e di ec ions in his ield. The s uc u e o he es o he pape is as ollows: Sec ion3 p esen s a lis o he mos ele an biclus e ing e iews. Sec ion4 g oups he p oblems ha HPC and Big Da a applica ions used by biclus e ing echniques mus add ess o suppo huge da ase s and ob ain hei esul s in he sho es possible ime. Biclus e ing echniques based on Big Da a and HPC applica ions a e examined in Sec ion 5. Sec ion 6 e iews he di e en biclus e alida ion me hods and alida ion p ocesses imple- men ed in biclus e ing echniques ha can alida e la ge amoun s o esul s in he sho es possible execu ion ime. Sec ion7 desc ibes he challenges ha biclus e ing and i s alida ion measu es mus ace o adap o a Big Da a ecosys em. Finally, Sec- ion8 summa ises he main conclusions de i ed om his wo k. 2 Re iew me hodology This s udy employs a s uc u ed p ocess based on he PRISMA 2020 p inciples o gua an ee scien i ic igou , anspa ency, and epea abili y. Two dis inc ye in e con- nec ed analyses we e conduc ed o encapsula e he ecen ad ancemen s in biclus e - ing me hodologies employed in bioin o ma ics, pa icula ly in ligh o Big Da a and high-pe o mance compu ing issues. The ini ial s udy concen a ed on disco e ing e iew a icles and su eys ha in es iga e biclus e ing echniques used in bioin o ma ics. The aim was o e alua e he dep h, b ead h, and emphasis o p e ious e alua ions, speci ically wi h hei han- dling o scalabili y, alida ion, and p ac ical use ulness in bioin o ma ics. The second s udy ocused on o iginal esea ch a icles ha p oposed o imple- men ed biclus e ing me hods inside High Pe o mance Compu ing (HPC) and Big Da a en i onmen s. This encompasses biclus e ing me hods using adi ional pa al- lel and dis ibu ed models (e.g. MPI, OpenMP), Big Da a amewo ks (e.g. Apache Spa k, Hadoop), and GPGPU accele a ion (e.g. CUDA, OpenCL). The objec i e is o delinea e ends, p oblems, and he echnological de iciencies in scalable biclus e deploymen s and hei applica ions wi hin he bioin o ma ics domain. 2.1 Sea ch s a egy andda a sou ces Bo h analyses employed sys ema ic sea ch s a egies ac oss h ee majo da abases: Scopus, PubMed, and Google Schola . The sea ches we e ca ied ou in May 2025 and we e limi ed o pee - e iewed a icles published be ween 1 Janua y 2009 and 25 May 2025. Only a icles w i en in English we e included. A.López-Fe nández e al. 1123 Page 6 o 52 Fo he e iew and su ey wo ks, que ies combined e ms such as: ”biclus e ing”, ”co-clus e ing”, ”su ey”, ” e iew”, ”bioin o ma ics”, and ”gene exp ession”. Fo he HPC/Big Da a implemen a ions, que ies included combina ions o : ”biclus e ing”, ”co-clus e ing”, ”pa allel compu ing”, ”dis ibu ed compu ing”, ”high-pe o mance compu ing”, ”HPC”, ”Big Da a”, ”Apache Spa k”, ”Hadoop”, ”GPU”, ”GPGPU”, ”CUDA”, and ”bioin o ma ics”. In all cases, Boolean ope a o s and pla o m-speci ic syn ax we e used. Al hough all analyses we e pe o med o all pla o ms, some illus a i e que ies o each da a- base a e shown below. PubMed que y (su ey and e iew iden i ica ion): ((biclus e ing[Ti le/Abs ac ] OR ”co- clus e ing”[Ti le/Abs ac ]) AND ( e iew[Ti le/ Abs ac ] OR su ey[Ti le/Abs ac ] OR benchma k*[Ti le/Abs ac ]) AND (bioin o ma ics[MeSH Te ms] OR ”gene exp ession”[Ti le/Abs ac ])) AND (”2009/01/01”[Da e - Publica ion] : ”3000”[Da e - Publica ion]) Scopus que y (Big Da a-based biclus e ing algo i hms): (TITLE-ABS-KEY(biclus e ing OR ”co-clus e ing”) AND TITLE-ABS-KEY(”Big Da a” OR ”Spa k” OR ”Apache Spa k” OR ”Hadoop” OR ”Apache Hadoop” OR ”MapReduce” OR ”Map-Reduce”) AND TITLE-ABS-KEY(bioin o ma ics OR ”gene exp ession”)) AND PUBYEAR> 2008 Google Schola que y (GPU-based biclus e ing algo i hms): ”biclus e ing” AND (”GPU” OR ”CUDA” OR ”GPGPU” OR ”OpenCL”) AND (bioin o ma ics OR ”gene exp ession”) - e iew -su ey a e :2008 As can be seen, each que y was ca e ully c a ed o ma ch he syn ax and index- ing s a egies o he co esponding pla o m. Ad anced il e s (e.g. publica ion yea ange, language, pee - e iewed jou nals) we e also applied when a ailable. 2.2 Eligibili y c i e ia andsc eening p ocess A icles we e managed using Zo e o, which enabled deduplica ion and s uc u ed agging. The selec ion was ca ied ou in wo s ages: (1) i le and abs ac sc eening and (2) ull- ex e alua ion. Sepa a e inclusion/exclusion c i e ia we e de ined o each analysis. Fo su ey analysis, s udies we e included i hey we e published e iews o su - eys ha analysed biclus e ing echniques applied speci ically o biological da a. P io i y was gi en o wo k ocused on bioin o ma ics o omics da ase s, pa icula ly hose in ol ing gene exp ession analysis. S udies we e excluded i hey did no ela e Biclus e ing inbioin o ma ics using big da a andHigh… Page 7 o 52 1123 o biological applica ions—such as e iews cen ed on ma ke ing o social ne wo k da a—o i hey lacked pee - e iewed s a us (e.g. edi o ials o in o mal o e iews). Fo he analysis o he HPC/Big Da a biclus e ing algo i hm analysis, a icles we e included i hey p oposed o applied biclus e ing echniques wi hin pa al- lel o dis ibu ed compu ing en i onmen s, including implemen a ions using GPU o GPGPU accele a ion. Wo ks ha desc ibed applica ions wi hin Big Da a ame- wo ks such as Apache Spa k o Hadoop we e also eligible. The s udies ha we e no included in his analysis we e hose ha did no ha e any compu a ional implemen- a ion, did no ocus on scalabili y o pe o mance, o only used simple oy da ase s ha we e no ele an o p ocessing la ge-scale biological da a. The i s analysis yielded 490 candida e e iew a icles, o which 24 me all inclu- sion c i e ia. The second analysis e ie ed 432 implemen a ion a icles, o which 24 we e selec ed a e ull- ex sc eening. 2.3 Da a ex ac ion andclassi ica ion Fo bo h analyses, me ada a we e ex ac ed, including publica ion yea , ci a ion coun ( ia Scopus/Google Schola ), jou nal qua ile, a ge o ganism (i applicable), da a ype (e.g. mic oa ay, RNA-Seq), compu a ional pla o m (e.g. Spa k, MPI, GPU), and e alua ion me ics. S udies om he second analysis we e u he ca ego ised by compu a ional pa adigm: • Pa allel and dis ibu ed compu ing (e.g. MPI, OpenMP) • Big Da a amewo ks (e.g. Spa k, Hadoop) • GPU-based app oaches (e.g. CUDA, OpenCL) This classi ica ion enabled a compa a i e syn hesis o algo i hmic s a egies and implemen a ion pa e ns. 2.4 Re iew scope and ype This wo k combines bo h sys ema ic and na a i e e iew elemen s. The sys em- a ic componen s p o ide a ep oducible s uc u e and empi ical ounda ion, while he na a i e analysis enables c i ical in e p e a ion o ends, bo lenecks, and me h- odological gaps. This hyb id app oach allows o bo h e idence-based syn hesis and o wa d-looking insigh . 2.5 E idence o in e na ional in e es and ele ance In bo h analyses, he a icles included co e mo e han 20 coun ies, wi h s ong ep- esen a ion om ins i u ions in he USA, China, Ge many, India, and Spain. App ox- ima ely 74% o he selec ed s udies we e published in jou nals anked in Q1/Q2 in hei ca ego y acco ding o he Scimago Jou nal Rank (SJR). I is also in e es ing o no e ha o hese publica ions anked in Q1, he selec ed su eys exceed he 80% A.López-Fe nández e al. 1123 Page 8 o 52 h eshold, while o HPC biclus e ing algo i hms hey s and a 67%. Ci a ion me ics also suppo he ele ance o his ield: he a e age ci a ion coun was 47.3, wi h a median o 31. These indica o s ein o ce he g owing in e na ional in e es in scal- able biclus e ing echniques in biomedical esea ch. 3 Rela ed wo k Nume ous biclus e ing e iews a e a ailable in he li e a u e, wi h he mos el- e an ones de ailed in Table2 o Appendix A. This able includes publica ion yea , au ho s, p incipal applica ion, and i s scope: biclus e ing echniques (me hods) and alida ion echniques ( alida ion). Re iews can span mul iple ca ego ies, as he g oups a e no mu ually exclusi e. Rega ding biclus e ing s a egies, S. Busygin e al. [20] o e an ex ensi e analysis o he ma hema ical p inciples in ol ed in he sea ch o biclus e s by means o a ious echniques ha a e classi ied based on hei applica ion domains. B. Pon es e al. [21] classi y algo i hms in o me ic and non-me ic g oups, employing di e se e alua ion me ics such as i e a i e g eedy sea ch, me a-heu is ics, clus e ing-based, p obabilis ic models, and linea algeb a. Recen wo ks ha e been published ha p o ide a mo e comp ehensi e analysis, such as he s udy by A. José-Ga cía e al. [22], which speci ically examines biclus e - ing algo i hms ha use me a-heu is ics as e alua ion me ics. Conce ning e iews o biclus e alida ion echniques, Tanay e al. [23] examine sco ing model in luence in he de ec ion o signi ican biclus e s, while San ama ia e al. [24] p opose me ics o illus a e he consis ency and e icacy o biclus e ing echniques o ge biological in o ma ion in he p esence o noise exposu e and o e lap be ween biclus e s. Al hough he a o emen ioned a icles p o ide heo e ical pe spec i es on biclus- e ing app oaches, hey ail o add ess he issues posed by biological da ase s and compu a ional obs acles. Thus, addi ional e alua ions seek o acqui e knowledge by conduc ing compa a i e analyses o a ious me hodologies. Fo example, E en e al. [25] ound ha he compu a ional e iciency o biclus e ing app oaches declines wi h la ge gene exp ession da ase s and mo e biclus e s. Addi ionally, no all me h- ods can gene a e biologically en iched biclus e s. Padilha e al.’s wo k [26] ei e - a ed he indings o E en e al. [25] and p o ided u he de ails on he ac o s ha in luence compu a ional e iciency, including noise in gene exp ession da a, biclus e o e lap, size, and quan i y. Also, K. Nicholls e al. [27] compa ed a ious biclus e - ing echniques, no ing ha adap i e algo i hms demons a e supe io compu a ional pe o mance. They also obse ed a co ela ion be ween noise, da ase size, biclus e coun , and dec eased compu a ional e iciency. Mo e ecen ly, Cas anho e al. [28] ha e o e ed ano he s udy in which hey in oduce a uni ied axonomy o biclus e - ing concep s and de ail a comp ehensi e analysis p ocess. Al hough hey iden i y h ee main implemen a ion s a egies o big da a biclus e ing algo i hms—pa allel compu ing, GPU-based execu ion, and MapReduce— hey limi he discussion o whe he each me hod is a no el p oposal o an adap a ion o exis ing algo i hms. Based on he highligh s om he a o emen ioned e iews, i can be de e mined ha biclus e ing e o s ha e long been ying o add ess he limi a ions and imp o e he quali y o he esul s. Howe e , e ining me a-heu is ics and add essing noise/ Biclus e ing inbioin o ma ics using big da a andHigh… Page 9 o 52 1123 o e lap challenges may no ully ex ac pe inen biological insigh s om as da ase s. Likewise, cu en biclus e alida ion me hods may lack adequacy when applied in Big Da a ecosys em. Rega ding hese conce ns, Xie e al. [6] p esen a compelling s udy on he impac o biclus e ing echniques in cu en biomedical con ex s. They highligh he lack o biclus e ing algo i hms ailo ed o la ge gene exp ession da ase s wi h high complexi y. The au ho s ad oca e o he de elopmen o no el biclus e ing me hods capable o managing biological and biomedical analy- ses wi hin Big Da a en i onmen s. Also, he e a e o he wo ks ha a e no e iews bu ha iden i y such p oblems. Fo example, in [29], he au ho s ocus on Big Da a biclus e ing app oaches using HPC amewo ks like Apache Hadoop and Spa k, along wi h pa allel sys ems wi h GPUs, o add ess scalabili y challenges. T adi ional biclus e ing me hods a e deemed imp ac ical o Big Da a asks due o hei inabil- i y o analyse da a e icien ly. The au ho s ou line key challenges, including le e ag- ing pa allelism, unde s anding ha dwa e limi a ions, and add essing he inc ease in biclus e quan i y and size wi h da a olume. They ecommend employing mul iple alida ion measu es when assessing biclus e ing pe o mance in noisy en i onmen s o wi h simila biclus e s. Addi ionally, hey cau ion agains o e -in e p e ing s a- is ical signi icance in gene en ichmen and pa hway analysis, ad oca ing o addi- ional il e ing o compa ison in Big Da a biclus e ing app oaches. As he analysis o cu en li e a u e e iews on biclus e ing echniques shows, many p oposals examine hese echniques om a ious pe spec i es. Howe e , none o hese wo ks add esses he challenges posed by he big da a en i onmen , such as p ocessing eno mous amoun s o da a and e alua ing a la ge numbe o biclus e s. This a icle aims o add ess his gap by p o iding a comp ehensi e s udy o biclus- e ing echniques om he pe spec i e o high-pe o mance compu ing (HPC). 3.1 C i ical analysis o p e ious wo k andiden i ied esea ch gaps Al hough a signi ican numbe o challenges ha e con ibu ed o he de elopmen o biclus e ing me hods o bioin o ma ics and Big Da a en i onmen s, se e al limi a ions and esea ch gaps pe sis . Fi s , many adi ional biclus e ing algo i hms exhibi limi ed scalabili y when applied o la ge-scale omics da ase s. These me h- ods a e o en implemen ed in sequen ial en i onmen s and lack e icien adap a- ions o dis ibu ed o pa allel compu ing amewo ks. Second, he in e p e abili y o he esul ing biclus e s is equen ly o e looked, which es ic s hei usabili y in biomedical con ex s whe e he explainabili y o compu a ional esul s is c ucial o domain expe s. Ano he ecu en issue is he absence o s anda dised e alua ion me hodologies; he di e si y o me ics and alida ion s a egies employed ac oss s udies hampe s ai compa isons and ep oducibili y. Mo eo e , he e is a gene al lack o in eg a ion o p io biological knowledge in o he biclus e ing p ocess. Gene on ology e ms, known pa hways, and egula- o y in e ac ions a e o en igno ed, despi e hei po en ial o guide mo e biologically meaning ul clus e ing. Fu he mo e, while mode n machine lea ning pa adigms such as deep lea ning and ensemble lea ning ha e shown success in o he bioin- o ma ic asks, hei applica ion in biclus e ing emains limi ed. The ep oducibili y A.López-Fe nández e al. 1123 Page 16 o 52 exp ession ma ix in o chunks co esponding o a ailable p ocesses. Each p o- cess is esponsible o cons uc ing local biclus e s om i s alloca ed da a chunk, while he main p ocess in eg a es hese biclus e s o p oduce he inal ou pu . As da a ansmission among p ocesses is a oided du ing local biclus e cons uc ion, he au ho s ind he communica ion cos o be negligible. Howe e , hey no e ha he me hod’s bo leneck a ises om cons uc ing biclus e s wi h a single p ima y p ocess, which can signi ican ly a ec compu ing pe o mance, especially wi h excep ionally la ge da ase s. To alida e hei app oach, he au ho s employ a syn he ic da ase measu ing 12651 x 30 o gene a e 250 biclus e s and claim simi- la esul s o a yeas gene exp ession da ase [76], hough no suppo ing da a is p o ided. Liu e al. [77] in oduce P-Biclus e , a biclus e ing me hod designed o cons uc - ing shi ing biclus e s [78] om gene exp ession da ase s con aining bo h con inu- ous and disc e e alues. The echnique ga he s shi pa e ns in o a 2x2 ma ix and u ilises he de ia ion om each pa e n o de e mine he shi end. Biclus e s a e hen gene a ed by inc emen ally adding ows and columns o each pa e n un il hei o se alue alls below a p ede ined h eshold. The au ho s e alua e hei app oach using eal gene exp ession a ays [79] wi hou noise con ol and compa e esul s wi h o he sequen ial me hods, assessing biclus e quali y based on de ia ion. How- e e , alida ion lacks suppo om echniques ensu ing he biological signi icance o biclus e s. In e ms o compu a ional p ope ies, P-Biclus e employs MPI o pa allel compu ing on CPU clus e s. The au ho s obse e imp o ed compu a ional pe o mance wi h se e al p ocesses up o se en bu no e ha pe o mance declines beyond his h eshold due o a lack o con ol o e dis ibu ed memo y a chi ec u e and p ocess communica ions. Hence, he me hod is ine ec i e o con inuous da a ans e o la ge da a olumes. Addi ionally, he s udy lacks de ails on pa allelised asks, CPU clus e u ilisa ion, synch onisa ion, and da a dis ibu ion, among o he aspec s. A. Nisa e al. [80] p esen a scalable and dis ibu ed algo i hm de i ed om Ahmad e  al.’s wo k [81], which e o mula es he Bipa i e Spec al Pa i ion- ing echnique [82] using g aph acing o achie e op imal solu ions. Thei me hod in ol es posi ioning each node o he bipa i e g aph a he ba ycen e o i s neigh- bou s o iden i y op imal solu ions. The biclus e ing p ocess consis s o wo pa allel asks: c osso e minimisa ion and biclus e iden i ica ion. Ini ially, he gene exp es- sion da ase is ho izon ally spli in o homogeneous chunks, each assigned o an a ailable p ocess. Da a no malisa ion is pe o med, ollowed by i e a i e ow (local) and column (global) eo de ing. Each p ocess conduc s a local sea ch o biclus- e s, and ep esen a i e biclus e s a e globally dis ibu ed. The au ho s employ MPI/ C++ o accele a e compu a ional pe o mance and analyse algo i hmic complex- i y and memo y a chi ec u e managemen . They no e ha MPI-based biclus e ing algo i hms incu communica ion cos s as da ase size g ows bu claim independence om ow coun by using c osso e minimisa ion. They demons a e scalabili y by analysing syn he ic da ase s wi h up o 20 million ows and 64 ixed columns and using up o 256 p ocesso s. Howe e , he s udy lacks analysis o eal gene exp es- sion da ase s and compa ison wi h o he me hods o compu a ional accele a ion echniques. Biclus e ing inbioin o ma ics using big da a andHigh… Page 17 o 52 1123 MFCM [83] is a biclus e ing echnique de i ed om he C-Means uzzy clus- e ing algo i hm [84] o gene a ing biclus e s om gene exp ession da ase s. The me hod u ilises Ma labMPI and an SPMD pa allel compu ing model o dis ibu e compu a ional wo kload ac oss mul iple p ocesso s. The da ase is di ided in o chunks co esponding o MATLAB p ocesses, ollowed by gene and sample clus e - ing o cons uc biclus e s. Gene cen es a e de e mined based on en opy a he han co ela ion coe icien s, wi h i e a ions con inuing un il a h eshold o ejec ed genes o a maximum i e a ion limi is eached. Real gene exp ession da ase s a e used, compa ing esul s o a sequen ial e sion in MATLAB. Compu a ional pe o mance imp o es a all s eps compa ed o he sequen ial e sion, bu challenges may a ise wi h inc eased p ocesses o la ge da ase s due o communica ion delays. The s udy lacks conside a ion o challenges in da a dis ibu ion, memo y a chi ec u e manage- men , and da a ans e s ac oss p ocesses, and i only e alua es up o eigh p ocesses, lea ing pe o mance unce ain y wi h mo e. The Bioconduc o unibic package [85] adap s he UniBic biclus e ing echnique [86] o pa allel compu ing o gene a e meaning ul biclus e s om gene exp ession da a. The modi ica ion aims o accommoda e high- h oughpu RNA-Seq, scRNA- Seq, and PCR da ase s, imp o ing un imes o la ge da ase s wi h hund eds o housands o columns and housands o ows. The au ho s p o ide de ailed compu- a ional decisions o pa allelising asks, excep o ma ix disc e isa ion. They e- implemen he algo i hm in C++11 wi h he OpenMP s anda d o enhanced pe - o mance. Res uc u ing he sou ce code op imises ea u es and add esses memo y segmen a ion issues. OpenMP is u ilised o asks like compu ing he Longes Com- mon Sequence (LCS) [87] be ween ow pai s, dis ibu ed ac oss CPU co es. How- e e , e idence on he algo i hm’s pe o mance wi h huge da ase s and he a ionale behind OpenMP usage is lacking. Consequen ly, he scalabili y and pe o mance o he OpenMP implemen a ion depend on single-machine ha dwa e. González-Dominguez e  al. [88] de eloped Pa BiBi , a pa allel e sion o he BiBi algo i hm [89], designed o ex ac biclus e s om bina y da ase s. Pa BiBi , implemen ed in C++11, uses MPI and POSIX h eads o exploi bo h sha ed and dis ibu ed memo y in a hyb id clus e en i onmen . Tasks wi h high compu a ional cos s, such as pa e n cons uc ion and ow agg ega ion, a e iden i ied. MPI and POSIX h eads dis ibu e he wo kload ac oss CPUs and nodes o enhance compu- a ional pe o mance. Howe e , da a ans e s be ween p ocesses may sa u a e he ne wo k, impac ing pe o mance. López e  al. [90] demons a ed Pa BiBi ’s deg- ada ion in pe o mance wi h dense da ase s and iden i ied memo y managemen conce ns o la ge da ase s. F aguela e al. [91] add essed hese issues wi h ScalaPa- BiBi , also using C++11, MPI, and POSIX Th eads. Sequen ial op imisa ions and me hodological weaks we e applied o enhance compu a ional speeds and suppo la ge da ase s. The main di e ence lies in ScalaPa BiBi ’s app oach o gene a ing pa e ns; i aims o dis ibu e pa e n se s o op imise biclus e elabo a ion. Howe e , noise sensi i i y in da ase s emains unadd essed ac oss all e sions. The COBRAC Con ex biclus e ing algo i hm [92] is capable o iden i ying po en ial biclus e s along wi h he associa ed ow and column clus e ing dend o- g ams. This algo i hm ope a es by educing he inpu da ase size and sol ing he con ex weigh ed biclus e ing p oblem. The u ilisa ion o con ex clus e ing ees A.López-Fe nández e al. 1123 Page 18 o 52 ensu es obus ness agains mino inpu a ia ions [93]. COBRAC is implemen ed in C/C++ o he algo i hm i sel and Py hon con aine s o b oade accessibili y. OpenMP is employed o accele a e he weigh ed con ex biclus e ing ask. While educing he inpu da ase size imp o es execu ion ime and enables handling la ge biological da ase s, u he compu a ional documen a ion is needed o elucida e hei esul s and esou ce usage. Addi ionally, he a ionale behind choosing OpenMP should be cla i ied, especially ega ding i s dependence on single-machine ha d- wa e. Al hough he au ho s p io i ise imp o ing usabili y, unning simula ions on a machine wi h 512 GB o RAM may no e lec ypical compu ing en i onmen s. Fu he explo a ion wi h la ge da ase s is wa an ed o e alua e he algo i hm’s scalabili y and pe o mance impac . The ARBic algo i hm de eloped by Ma e al. [94] is an accele a ed ule-based biclus e ing algo i hm designed o iden i y ele an gene condi ion pa e ns in bio- medical da ase s. Implemen ed in C++, ARBic employs OpenMP o pa allelise he expansion phase o seed biclus e s, which is compu a ionally in ensi e due o he high dimensionali y o omics da a. OpenMP enables ARBic o exploi sha ed memo y pa allelism by dis ibu ing he p ocessing o candida e seeds ac oss mul- iple h eads. This pa allel s a egy signi ican ly educes execu ion ime du ing he explo a ion o he biclus e sea ch space. Howe e , he use o OpenMP con ines he algo i hm o sha ed memo y a chi ec u es, limi ing scalabili y o la ge da ase s ha exceed he memo y capaci y o a single node. Fu he mo e, he inse ion o esul s in o sha ed s uc u es equi es c i ical sec ions, which may in oduce synch onisa- ion o e head and limi pa allel e iciency. To enhance ARBic’s pe o mance, u u e e sions could adop hyb id pa allelism by in eg a ing OpenMP wi h MPI o ask- based un ime sys ems, allowing dis ibu ed memo y suppo and mo e e ec i e wo kload balancing. Addi ionally, op imising memo y access pa e ns and educing con en ion in sha ed da a s uc u es could u he imp o e scalabili y and pa allel h oughpu . B. Ba uah e al. [95] in oduce EnsemBic, an assembly-based biclus e ing me hod ha in eg a es se e al me hods (Laplace P io , iBBiG, and xMo i ) o ex ac unc- ionally ele an biclus e s om gene exp ession da a. The me hod iden i ies high- quali y biclus e s using p- alues de i ed om FuncAssocia e 3.0 and employs an i e a i e app oach o elimina e gene ic simila i ies wi hin biclus e s. The au ho s emphasise ha he design acili a es he execu ion o base algo i hms in pa allel, po en ially dec easing compu a ion ime; howe e , he R implemen a ion does no explici ly inco po a e pa allelisa ion echniques such as OpenMP, MPI, o na i e R lib a ies o pa allel compu ing. This seeming lexibili y indica es ha pa allelisa- ion is ei he ex e nally managed o subjec o use implemen a ion. EnsemBic has been assessed using ou au hen ic da a se s ( wo mic oa ays and wo RNA-seq), demons a ing consis en enhancemen s o e s andalone me hods in bo h opologi- cal me ics (in e nal densi y, modula i y, ODF) and biological me ics (GO en ich- men , pa hway analysis, and alida ion by ChIP-seq). Mo eo e , he inclusion o he p- alue as a quali y c i e ion ende s he app oach sensi i e o non-signi ican pa - e ns and enhances i s obus ness agains noise. The algo i hm unc ions on eal con- inuous da a; howe e , i s use wi h syn he ic, discon inuous, o bina y da a has no been documen ed. Biclus e ing inbioin o ma ics using big da a andHigh… Page 19 o 52 1123 The e a e addi ional biclus e ing echniques ha a e accele a ed using con en- ional pa allel models, bu insu icien in o ma ion is a ailable due o he una ail- abili y o so wa e o a lack o desc ip ion ega ding he pa allel model and com- pu a ional esou ces. One such example is PBD-SPEA2 [96], an adap a ion o he SPEA2 mul i-objec i e e olu iona y algo i hm [97] o gene exp ession da a. PBD-SPEA2 uses a ixed-leng h, in ege -based encoding o e alua ing biclus e se s. Objec i e unc ions include MSR [14], BSize [22], and VAR [98], wi h he Cheng–Chu ch [14] heu is ic o p oducing new solu ions. Pa allel compu ing is applied o he dynamic coding scheme ask, al hough speci ic pa allelisa ion me h- ods, s anda ds, o models a e no p o ided, no is he algo i hm’s sou ce code a ail- able, hinde ing u he explo a ion o compu a ional ea u es. In MPI-based biclus e ing algo i hms, he size and di e si y o inpu da ase s play a c ucial ole in speeding up compu ing pe o mance. La ge da ase s o en lead o memo y managemen and da a ans e challenges due o communica ion sa u a ion among memo y egions ac oss p ocesso s in he clus e . ScalaPa BiBi o e s a solu- ion o hese limi a ions. On he o he hand, biclus e ing algo i hms using OpenMP ace limi a ions inhe en in his implemen a ion, which p ima ily pa allelises p o- g ams in a sha ed memo y en i onmen , a oiding communica ion o e head bu es ic ing hem o a single compu e ’s ha dwa e esou ces. The COBRAC algo i hm exempli ies hese cons ain s by comp essing da ase size o educe execu ion ime compa ed o i s sequen ial e sion. Addi ionally, conside a ions such as ask selec- ion o pa allelisa ion, a chi ec u e, and memo y managemen mus be add essed. While algo i hms may ace limi a ions due o ha dwa e esou ces, such an app oach ensu es scalabili y wi hou o e loading o memo y o e low. Fu he mo e, as da a- se olume inc eases, he likelihood o ob aining mo e biclus e s ises, equi ing ca e ul conside a ion du ing p ocessing and memo y managemen . 5.2 Pa allel biclus e ing wi hbig da a p og amming pa adigms The eme gence o Big Da a p og amming pa adigms add esses he limi a ions o adi ional pa allel and dis ibu ed models in handling la ge da ase s e icien ly. Con- sequen ly, p og amming pa adigms like MapReduce, along wi h pla o ms such as Apache Hadoop, Apache Spa k, and MATLAB, o e solu ions o massi ely pa - allel o dis ibu ed biclus e ing. Two ables p esen he cha ac e is ics o he ana- lysed biclus e ing me hods o you e e ence. Table6 o Appendix A ou lines he cha ac e is ics o biclus e ing echniques ailo ed o a Big Da a en i onmen . The i s column lis s he names o he biclus e ing algo i hm o au ho s, i no speci ied. The second column indica es he Big Da a pla o m used. The ypes o da ase s used in expe imen s a e de ailed in he hi d column: 0 o syn he ic da ase s, 1 o eal da ase s, and 2 o a combina ion o bo h. The ou h column deno es whe he com- pa isons we e made wi h o he s udies: 0 o no compa ison, 1 o compa isons wi h sequen ial me hods, and 2 o compa isons wi h o he Big Da a-adjus ed biclus e - ing algo i hms o deploymen on di e en Big Da a pla o ms. The i h column no es he algo i hm’s abili y o handle noise. Da a ypes used in expe imen s a e A.López-Fe nández e al. 1123 Page 20 o 52 lis ed in he six h column, acknowledging ha addi ional da a ypes migh also be compa ible. Table7 o Appendix A delinea es he compu a ional cha ac e is ics o he dis- cussed biclus e ing me hods. The i s column p esen s he algo i hm’s name o au ho s’ names. The second column indica es he numbe o MR jobs equi ed o gene a ing esul s, whe e 0 signi ies a small numbe and 1 indica es a highe num- be . The hi d column deno es he me hod’s capabili y o da a spli ing, c ucial o balancing wo kload and da a ans e s [99]. The ou h column indica es whe he me hods ha e s a egies o minimise I/O ope a ions, which is c i ical o pe o - mance, especially in Apache Hadoop se ups. The i h column speci ies i in e medi- a e da a is s o ed in main memo y o in-memo y compu a ion, educing disk I/O. The six h column iden i ies whe he MR jobs a e asynch onous, impac ing ha dwa e and so wa e esou ce u ilisa ion. The se en h column e alua es he in es iga ion o op imisa ion s a egies, which a e c ucial o e icien pa allelism. The inal column indica es he ype o scalabili y expe imen s conduc ed: 1 o inc easing da a ol- ume, 2 o aising he numbe o clus e nodes, and 3 i bo h scalabili y es s we e pe o med. BiTM-MR [100] is an ea ly biclus e ing app oach using he MapReduce pa a- digm, implemen ed wi h Apache Spa k o ex ac ing biclus e s om massi e da a- se s. The me hod ans o ms da a ma ices in o a block s uc u e o ganised as a opo- logical map and employs wo bina y ma ices o alida e ow–column ela ionships. O e coming challenges like da a loading, ailu e sa e y, and algo i hm design, he au ho s use Apache Spa k o aul co ec ion, da a dis ibu ion, and managemen . The me hod uses wo MR jobs o i e a e ows and columns and adjus pa ame e s. The au ho s signi ican ly imp o e compu a ional e iciency by alloca ing independ- en MR jobs o ows and columns, minimising I/O ope a ions, and enabling asyn- ch onous p ocessing. Howe e , de eloping a da a pa i ioning echnique such ha each MR job akes ca e o a g oup o homogeneous ows and columns is an addi- ional compu ing ac o ha he au ho s ha e no accoun ed o and ha could be use ul o enhancing hei algo i hm design. Syn he ic da ase s wi h up o wo mil- lion ows demons a e nea -ideal compu a ional e iciency as da ase size and clus e co es inc ease. Howe e , compa isons wi h o he biclus e ing algo i hms o pe o - mance accele a ion esou ces would ha e added alue. The au ho s no e ha Apache Spa k ini ialisa ion impac s pe o mance mo e on smalle da ase s. Ruiqi e al. [101] adap ed he non-nega i e ma ix ac o isa ion biclus e ing algo- i hm [102] using Apache Hadoop o handle 2D spa se ma ices up o one million by one million. Thei app oach equi es a non-nega i e ma ix and a epe i ion coun as inpu , using i e MR jobs pe i e a ion o upda e each ou pu subma ix. Howe e , his app oach’s hea y eliance on MR jobs may impac compu a ional pe o mance on la ge da ase s due o equen disk access cos s. Conside a ions such as educ- ing I/O ope a ions, op imising da a pa i ioning, and enabling asynch onous com- munica ion be ween map and educe unc ions we e no add essed. The s udy sug- ges s oom o signi ican imp o emen in bo h compu a ional cos and scalabili y. An e alua ion wi h syn he ic da ase s and a da ase exceeding one million ows and columns om he STRING da abase [103] was conduc ed, analysing compu a ional e iciency conce ning he numbe o nonze o in e ac ions. Howe e , he e was no Biclus e ing inbioin o ma ics using big da a andHigh… Page 21 o 52 1123 analysis o how well i scales wi h mo e Hadoop clus e nodes o how i compa es o o he biclus e ing me hods o di e en esou ces o handling la ge da ase s. MR-GABiT [104] is designed o gene ic biclus e ing o ime se ies da ase s om mic oa ay expe imen s. Despi e he da ase s being h ee-dimensional, he au ho s apply biclus e ing o ex ac local pa e ns a di e en ime poin s, which a e hen compa ed o de e mine op imal pa e ns. Using he MATLAB MapReduce ame- wo k, hey add ess dimensionali y and scalabili y issues. Da a pa i ioning by ime poin s educes MR job coun , imp o ing compu a ional pe o mance. Each MR ask employs a map unc ion o ex ac he bes indi idual om da a chunks, wi h educe unc ions combining hese bes indi iduals. Howe e , limi a ions include he inabil- i y o educe disk I/O ope a ions and educe unc ions wai ing o all map unc ions o comple e. E alua ion wi h yeas cell cycle ime se ies da ase s [105] is conduc ed, bu he scalabili y wi h inc eased clus e nodes is no es ed, and he e is no com- pa ison wi h o he gene ic biclus e ing me hods o Big Da a esou ces. The e a e addi ional biclus e ing s udies using he MapReduce pa adigm whe e compu a ional de ails a e insu icien and he so wa e is una ailable. One such wo k is MCC [106], an adap a ion o he Cheng–Chu ch (CC) biclus e ing echnique [14]. MCC andomly gene a es nume ous unique suba ays and execu es mo e compu a- ions on hem han he o iginal CC algo i hm, making i compu a ionally in ensi e. While MapReduce is used, speci ics on i s adap a ion, da a pa i ioning, and dis i- bu ion a e no p o ided. Real gene exp ession da ase s show ha MCC wo ks be - e han he sequen ial CC algo i hm in e ms o accu acy and speed. The au ho s acknowledge MCC’s ine iciency in memo y managemen due o he need o s o e nume ous in e media e a ays and he high numbe o I/O ope a ions. As can be seen, his subsec ion highligh s he sca ci y o bioin o ma ics- ela ed biclus e ing algo i hms on Big Da a pla o ms like Spa k o Hadoop. Consequen ly, his subsec ion also includes biclus e ing algo i hms ha ha e been c ea ed in disci- plines o science un ela ed o bioin o ma ics bu ha a e based on he MapReduce pa adigm. Then, i is possible o c ea e ais as o whe he he lack o algo i hms is due o he complexi y o he biclus e ing echnique’s de ini ion o o scien i ic ield- speci ic ac o s in bioin o ma ics. DisCo [107] ea anges subma ix ows and col- umns o mee a quali y h eshold, using key– alue pai s o s o e ows. Bha naga e al. [108] de eloped a MapReduce biclus e ing echnique using o mal concep analysis (FCA) o bina y da ase s [109]. Thei ocus is on imp o ing compu a ional pe o mance as da ase size inc eases, al hough hey ha e no explo ed scalabili y wi h node numbe s, and he BiBi sequen ial biclus e ing me hod achie es as e execu ion speeds in eigh o he ou een o e all pe o mance e alua ions. Lin e al. [4] de eloped biclus e ing algo i hms o he elecommunica ions ield, wi h e - sions o Spa k (SP-PLSS) and Hadoop (MR-PLSS). They aim o iden i y p o i able cus ome s wi h simila buying beha iou s, esul ing in be e e iciency and scalabil- i y wi h Apache Spa k o la ge da ase s. This analysis demons a es ha , ega dless o he ield o s udy, biclus e ing echniques ha e no been widely implemen ed on Big Da a pla o ms such as Apache Spa k and Apache Hadoop. In addi ion, he e is a signi ican gap be ween he p esen numbe o biclus e ing algo i hms p oduced on hese Big Da a pla o ms and o he ypes o compu a ional accele a ion applica ions. As a esul , mul iple hypo heses may eme ge om di e se iewpoin s. A.López-Fe nández e al. 1123 Page 22 o 52 Many adi ional biclus e ing me hods need o cons an ly p ocess one pa o he da ase , which makes i di icul o design hese me hods [14, 89, 94, 110, 111]. Consequen ly, no all s a egies easily adap o he MapReduce pa adigm due o his complexi y. The algo i hm design mus conside ac o s like MR job coun , da a- se pa i ioning, minimising I/O ope a ions, in-memo y sys ems, and asynch onous communica ion be ween map and educe unc ions. P ocessing la ge da ase s can g ea ly enhance pe o mance, bu he di e ences in bioin o ma ics da ase s migh make i ha d o hese algo i hms o wo k well, especially when he e a e only a ew genes and expe imen al condi ions [108]. Also, Big Da a pla o m ini ialisa ion ime impac s pe o mance [49, 100], and issues like educe unc ions wai ing o map unc ion comple ion and uncon olled execu ion o de can deg ade pe o mance. Biclus e ing algo i hms on Apache Spa k ou pe o m hose on Apache Hadoop due o in e media e I/O ope a ions elying on memo y a he han disk [4]. 5.3 Pa allel biclus e ing onGPUs The GPGPU concep ha nesses he p ocessing esou ces o GPU de ices o accel- e a e compu a ional pe o mance, making hem iable al e na i es o supplemen s o CPU p ocessing o biclus e ing algo i hms [112]. Howe e , biclus e ing algo- i hms ailo ed o GPGPU mus conside a ious ac o s ou lined in subsec ion4.3 o op imise compu a ional e iciency and suppo Big Da a da ase s. These ac o s a e analysed o each biclus e ing algo i hm adap ed o GPGPU and p esen ed in wo ables. Table8 o Appendix A p o ides key ea u es o biclus e ing echniques implemen ed on GPGPU. The i s column lis s algo i hm names, while he second column indica es da ase ypes: 0 o syn he ic, 1 o eal, and 2 o bo h. The hi d column indica es i compa isons we e made wi h o he me hods: 1 o sequen ial me hods and 2 o o he GPGPU-adap ed algo i hms. The ou h column no es i he app oach handles da ase noise. The i h column speci ies he da a ypes used, acknowledging ha o he ypes may also be applicable. The inal columns de ail suppo ed da a ypes o each algo i hm. Table9 o Appendix A de ails he compu a ional aspec s o biclus e ing algo- i hms. The i s column lis s algo i hm names. The second column speci ies mul i- GPU suppo and he pa allelism echnique. Sha ed memo y usage is no ed in he hi d column. The ou h column indica es he use o coalescence. The memo y allo- ca ion me hod is speci ied in he i h column (0 o non-pinned, o he wise pinned). The da a ans e me hod (global memo y o uni ied memo y) is desc ibed in he six h column. Occupancy le els, e lec ing esou ce u ilisa ion, a e in he se en h column. To be e e lec he compu a ional oo p in o he su eyed ools, we clas- si ied esou ce occupancy in o h ee ca ego ies based on es ima ed a e age usage o compu a ional esou ces: low (<40%), medium (40–70%), and high (>70%). The inal column indica es how synch onisa ion poin s a e used. A nedo-Fe nández e al. [113] in oduced an adap a ion o he FLOC biclus e - ing me hod [114], ocusing on he compu a ional in ensi y o calcula ing he MSR measu e [14] o biclus e alida ion. U ilising CUDA o his ask, hey demons a e imp o ed compu a ional pe o mance compa ed o he sequen ial e sion, using up Biclus e ing inbioin o ma ics using big da a andHigh… Page 23 o 52 1123 o 2000 x 2000 squa e syn he ic da ase s. Howe e , hey did no ully exploi he GPU’s capabili ies, no ing da a ans e ia he PCI-Exp ess bus as a bo leneck and no conside ing mul i-GPU wo kload spli ing. The au ho s also discussed he impac o h ead coun pe CUDA block on compu a ional e iciency, obse ing diminishing e u ns wi h excessi ely high h ead coun s. Rega ding memo y man- agemen , he au ho s do no speci y he a ious o ms o memo y p esen in a GPU de ice. The e o e, i is assumed ha da a is always used om he GPU’s global memo y. Liu e al. [17] pa allelised he ask o p ocessing column pai s in he Geome - ic Biclus e ing algo i hm (GBC) [115]. They de eloped h ee pa allel implemen- a ions: one using POSIX Th eads, ano he using CUDA, and a hi d employing a ield-p og ammable ga e a ay (FPGA) [116]. In he CUDA e sion, columns a e pai ed pe CUDA block, wi h each h ead esponsible o combining a po ion o each column using sha ed memo y o ensu e high coalescence and a oid memo y con lic s. Real da ase expe imen s indica e ha he GPU e sion achie es he high- es speedup, while he FPGA e sion is he mos ene gy-e icien . The au ho s dis- co e ha minimising da a ans e s be ween RAM and GPU global memo y is c u- cial, especially o la ge da ase s in he GPU e sion. In ano he wo k, Liu e al. [117] apply he GBC algo i hm o iden i y neu al p ocessing pa e ns in mic oa ay da ase s, p esen ing h ee CUDA e sions wi h dis inc op imisa ions. The i s e - sion uses sha ed memo y o s o e column pai chunks, ensu ing no o e head ega d- less o chunk size. A e loading he a ge column pai in o sha ed memo y, me ging ope a ions a e conduc ed, and ou comes a e s o ed in global memo y. Mo eo e , he chunking o column pai s p e en s he sha ed memo y om o e loading, necessi a - ing a global synch onisa ion o e e y p ocessed chunk. Acco ding o he au ho s, his global synch onisa ion incu s a cos by es o ing ine icien column pai s in he global memo y. The second e sion a emp s ope a ions on mul iple column com- bina ions o mi iga e his, while he hi d e sion ocuses on educing index upda e ime. In o de o accomplish his, hey examine he e ec o con inuous ans e s be ween GPU de ices and he CPU, a p oblem iden i ied in p io wo k [17]. Real da ase expe imen s show he second e sion achie es he bes speedup, emphasis- ing he impo ance o da a euse, minimising global memo y accesses, and dis ib- u ing wo kloads be ween CPU and GPU o la ge da ase s. Despi e hese op imisa- ions, none o hese e sions implemen s a mul i-GPU design, which could u he enhance wo kload alloca ion and suppo o la ge da ase s. Mejía-Roa e al. [118] in oduced a mul i-GPU adap a ion o he non-nega i e ma ix ac o isa ion (NMF) [102] algo i hm, using he MPI s anda d o mul i-GPU synch onisa ion. The algo i hm pa i ions he da ase in o blocks and dis ibu es i ac oss GPUs, subsequen ly decomposing he da a in global memo y. A 1D block se up achie es coalescence by accessing consecu i e memo y add esses. Asynch o- nous CPU- o-GPU da a ans e s enhance compu a ional pe o mance. Real da ase expe imen s con i m ha he GPU e sion ou pe o ms he sequen ial CPU-based algo i hm, wi h speedup inc easing wi h da ase size. Scalabili y analysis e eals ha ac o s like da a ans e s, synch onisa ion o e heads, and da ase size a e c i i- cal in mul i-GPU designs [119]. Fo la ge da ase s, he mul i-GPU app oach is p e - e able, while a single GPU su ices o smalle ones. A.López-Fe nández e al. 1123 Page 24 o 52 The Condi ion-dependen Co ela ion Subg oups (CCS) [120] algo i hm, de el- oped in CUDA, ully u ilises GPU de ices o biclus e gene a ion. Real and syn- he ic da ase s demons a e a 20x speedup compa ed o he sequen ial implemen- a ion. Howe e , unlike p e ious wo ks, he algo i hm does no employ da ase chunking using CUDA block/ h ead decomposi ion, which limi s i s scalabili y o he GPU memo y size. Sha ed memo y expedi es biclus e c ea ion bu may o e - load wi h excessi e columns, capped a 200. Each CUDA block cons uc s a biclus- e using a single h ead, esul ing in low occupancy. The me hod lacks suppo o mul i-GPU a chi ec u es; hus, i does no explo e wo kload dis ibu ion o scalabil- i y o la ge da ase s. EBIC, de eloped by O zechowski e al. [121], is a pa allel e olu iona y biclus e - ing me hod designed o disco e nume ous ele an pa e ns wi h high p ecision. I uses a mul i-GPU a chi ec u e o dis ibu e da ase ows ac oss GPUs, handling he en i e biclus e gene a ion p ocess. The au ho s op imise wo kload dis ibu ion ac oss GPUs, CUDA blocks, and h eads by using sha ed memo y o compu ing i ness unc ions and gene a ing biclus e s. Expe imen al esul s wi h syn he ic and eal da ase s demons a e EBIC’s compu a ional e iciency, high coalescence, and occupancy. EBIC ou pe o ms al e na i e echniques like CCS by up o 12 imes o la ge da ase s. Howe e , he algo i hm’s scalabili y o Big Da a is limi ed by he maximum suppo o 60000 ows pe GPU de ice. González-Domínguez e al. [122] implemen ed CUBiBi , a mul i-GPU e sion o he BiBi [89] biclus e ing algo i hm. CUBiBi employs GPU de ices o add ows o po en ial biclus e s, wi h he emaining p ocessing done on he CPU. In he case o a mul i-co e p ocesso , i can exploi he mul iple CPU co es while inding po en ial biclus e s hanks o a pa allelisa ion wi h POSIX h eads. CuBiBi employs he GPU’s sha ed memo y o s o e seeds o compu a ional pe o mance. Howe e , his decision limi s scalabili y o la ge da ase s because his memo y is iny and he numbe o pa e ns inc eases as he da ase g ows. Fu he mo e, he o iginal me hod ou pe o ms i o small da ase s. gBiBi [90], ano he e sion o BiBi , uses a mul i- GPU a chi ec u e o p ocess massi e bina y da ase s e icien ly. The me hodology add esses da a ans e , wo kload dis ibu ion, and esou ce u ilisa ion by ou pe - o ming o he adap a ions and being he only modi ied e sion capable o handling big da ase s. In ano he wo k, he au ho s c ea ed a Py hon package named bioSci- ence ha uses HPC o speed up a ious da a mining echniques, such as he BiBi algo i hm, using CPU and mul i-GPU clus e s [123]. The u ilisa ion o GPU de ices o accele a ing biclus e ing algo i hms has become a p e alen end, wi h CUDA being he mos used pla o m o his pu - pose. Va ious aspec s need conside a ion when de eloping and op imising biclus- e ing algo i hms in a mul i-GPU en i onmen , as obse ed in he p e ious wo ks. While GPGPU pa allel algo i hms gene ally ou pe o m sequen ial ones in e ms o compu ing pe o mance, many s uggle o handle massi e da ase s e ec i ely. In he majo i y o he analysed algo i hms, au ho s s ess he impo ance o plan- ning esou ce alloca ion o GPU u ilisa ion. CCS, o ins ance, employs a CUDA block o pa allelise a ask, whe eas CUBiBi , NMF, EBIC, and o he algo i hms use he h eads o a CUDA block o do he pa allelisa ion ask. This decision is di ec ly dependen on he algo i hm’s occupancy a e and, consequen ly, on whe he he Biclus e ing inbioin o ma ics using big da a andHigh… Page 25 o 52 1123 implemen a ion will use he ull powe p o ided by he g id o each GPU de ice. Fo algo i hms suppo ing mul i-GPU a chi ec u es, wo kload dis ibu ion among de ices is ca e ully conside ed. These wo ks use pa allelisa ion echniques like POSIX Th eads, MPI, and OpenMP o dis ibu e asks, bu he mos e ec i e s a - egy emains unclea due o he a ie y o op ions a ailable. Rega ding memo y- ela ed elemen s, he au ho s o se e al wo ks iden i y CPU–GPU da a ans e as a pe o mance bo leneck due o communica ion bus lim- i a ions, leading some o minimise ans e s. Con inuous global memo y access can also impede e iciency, p omp ing he use o sha ed memo y o boos speed. How- e e , imp ope sha ed memo y use, as seen in CuBiBi o CCS, can hinde pe o - mance on la ge da ase s o smalle da ase s. Thus, sha ed memo y usage’s e icacy a ies wi h da ase size and should be ca e ully assessed. Mos algo i hms employ ixed memo y ese a ion and alloca ion s a egies. Synch onisa ion be ween GPU de ices is c ucial in mul i-GPU sys ems, bu i a ies based on algo i hm echnique. Fo example, NMF and gBiBi use asynch onous communica ions o minimise syn- ch onisa ion and imp o e e iciency. 5.4 Pe o mance conside a ions ingene exp ession da a analysis Implemen ing biclus e ing algo i hms using HPC echnologies in big da a en i on- men s has shown a iable pe o mance depending on he speci ic cha ac e is ics o he biological ask. In he con ex o gene exp ession da a analysis, which ypically in ol es la ge-scale, high-dimensional, and spa se ma ices, each compu a ional model p esen s dis inc ad an ages and limi a ions. Benchma k da ase s a e com- monly u ilised o assess he pe o mance and scalabili y o HPC biclus e ing me h- ods wi hin a Big Da a amewo k, ensu ing bo h e icacy and biological signi icance. The Cance Genome A las (TCGA) and he Geno ype-Tissue Exp ession (GTEX) p ojec [124] a e he mos equen ly u ilised esou ces. The TCGA o e s ex ensi e, high-dimensional gene exp ession da a ac oss many cance ypes, se ing as a alu- able esou ce o assessing he scalabili y and p ecision o biclus e ing algo i hms in eal-wo ld biomedical con ex s. Con e sely, GTEX p o ides comp ehensi e ansc ip ome da a om se e al non-diseased human issues, acili a ing he iden- i ica ion o issue-speci ic gene co-exp ession modules. The ex ensi e u ilisa ion o TCGA and GTEX s ems om hei public accessibili y and comp ehensi eness, as well as he chance hey p o ide o e alua e algo i hmic esilience in high- h oughpu se ings, which is c ucial o implemen ing biclus e ing app oaches wi hin HPC o Big Da a amewo ks. Syn he ic da ase s a e u ilised in a ious esea ch o e alua e algo i hm pe o mance unde con olled se ings o noise, size, and s uc u e. Table 1 p o ides a compa a i e summa y o he main HPC pa adigms dis- cussed in his sec ion, ou lining hei associa ed echnologies, s eng hs, limi a- ions, and ypical applica ions in gene exp ession biclus e ing asks. T adi ional pa allel and dis ibu ed models, such as MPI-based clus e s, a e p o icien a managing la ge-scale gene exp ession da ase s, especially in jobs cha ac e ised by signi ican da a pa allelism, such as gene o sample il e ing. Ne e heless, hei e icacy may diminish in i e a i e biclus e ing con ex s due o in e -node A.López-Fe nández e al. 1123 Page 32 o 52 Based on he esea ch conduc ed in he p eceding sec ions o his wo k, he p i- ma y challenges ha biclus e ing mus o e come in o de o adap o a Big Da a ecosys em a e ou lined below. These challenges a e ca ego ised in o ou p ima y g oups: da ase s, biclus e ing app oaches, biclus e ing alida ion app oaches, and isualisa ion and in e p e a ion o he esul s. 7.1 Da a‑cen ic challenges Due o he exponen ial g ow h o biological and biomedical da a alongside ad ance- men s in NGS [166] echnologies, bioin o ma ics is con on ed wi h signi ican challenges in s o ing and analysing as da ase s. The pace o da a gene a ion om sequencing ou paces he compu a ional esou ces’ capaci y o p ocess such la ge olumes, leading o an expanding gap be ween hem [167]. Nume ous p ojec s and eposi o ies a e eme ging o s o e inc easingly massi e and in ica e da ase s. Fo ins ance, The Cance Genome A las (TCGA) [18] amassed 2.5 pe aby es o di e se biological da a, including mRNA, miRNA, and p o ein exp ession da a, along wi h his ology slides and gene ic a ia ion da a. The ENCODE p ojec [168] ocuses on anno a ing unc ional sequences in he human genome, expanding o include da a om o he o ganisms like mice, lies, and wo ms, o alling o e i e e aby es. The Eu opean Bioin o ma ics Ins i u e (EBI) [169] s o es mo e han 390 pe a- by es o aw biological da a, making i one o he la ges eposi o ies globally. As Biclus e ing Pipeline in Big Da a Bioin o ma ics wi h HPC In eg a ion Da a Sou cesGenomic Da a, Gene Exp ession Ma ices (HDFS/Cloud S o age) P ep ocessingNo maliza ion, Fea u e Selec ion(Spa k + GPU) Biclus e ing Algo i hmsDis ibu ed FLOCK, Plaid(MPI + CUDA) Visualiza ionHea maps, D3.js, Tableau Applica ionsD ug Disco e y, Mul i-Omics Big Da a HPC DATA SOURCES DATA PREPROCESING BICLUSTERING VISUALIZATION Valida ion & E alua ionGO En ichmen , Scalabili y Me ics VALIDATION APPLICATIONS Fig. 3 Bioin o ma ics Big Da a biclus e ing pipeline wi h HPC in eg a ion (MPI/GPU), designed o scalabili y in applica ions such as d ug disco e y and mul i-omics in eg a ion Biclus e ing inbioin o ma ics using big da a andHigh… Page 33 o 52 1123 hese eposi o ies g ow in olume, complexi y, and di e si y, ex ac ing meaning ul insigh s becomes inc easingly challenging. The c ea ion and p ep ocessing o da ase s om NGS pla o ms can impac da a p ecision and quali y due o ac o s like low-quali y eads, duplica e eads, o inse - ions/dele ions [170]. Reads ep esen he sequenced base pai s (bp) om DNA ag- men s. Da a ans o ma ion, such as om FASTQ o FASTA, is c ucial o e i ying da a quali y and elimina ing noise [171]. Howe e , he p oduc ion and p ep ocessing o la ge da ase s a e ime-consuming p ocesses. To add ess hese challenges, he sci- en i ic communi y is de eloping Big Da a and HPC ools using dis ibu ed memo y sys ems [172–174]. The con inuous gene a ion o biological and biomedical da a p esen s challenges in s o age and managemen . Cloud compu ing has eme ged as an e ec i e solu ion, as e idenced by a ious s udies [175, 176]. Despi e i s bene i s, e o s a e unde - way o de elop da a comp ession echniques o educe cloud compu ing cos s [177]. La ency is ano he conce n, as da a e ie al om he cloud o scien i ic da a analy- sis can be ime-consuming. Shi ing da a p ocessing o he cloud can add ess his issue, educing la ency and cos s and accele a ing esul gene a ion. Amazon AWS, o example, o e s he Amazon AWS Genomics se ice, equipped wi h ools o p ocess la ge olumes o genomic da a e icien ly [178, 179]. Cloud-based s o age equi es conduc ing da a analysis compu a ions in he cloud o mi iga e la ency in esul gene a ion. Scalable ools like SeqPig [180], BigBWA [181], GMQL [182], SeQuiLa-co [183], and SeQual [184] ensu e da a in eg i y du ing quali y con ol and p ep ocessing, especially when dealing wi h ex ensi e aw sequencing da a. 7.2 Algo i hmic challenges T adi ional biclus e ing me hods ha e p ima ily ocused on imp o ing esul qual- i y, bu hey o en s uggle wi h handling la ge da ase s, esul ing in educed use ul- ness. Challenges in p oducing biclus e s om ex ensi e da ase s, in luenced by a i- ous ac o s such as applica ion domains, ypes, s uc u es, o dimensions, con ibu e o his issue [6, 146]. Addi ionally, he cha ac e is ics o la ge biological da ase s inc ease he likelihood o gene a ing la ge and mo e nume ous biclus e s, equi ing he use o Big Da a and HPC ools o expedi ed biclus e gene a ion. To choose he app op ia e Big Da a o HPC applica ion o designing a biclus- e ing me hodology, unde s anding he p oblem’s na u e, pa allelisable asks, and da ase cha ac e is ics is c ucial. Fo asks equi ing high p ocessing capabili y wi h la ge - han-usual bu no massi e da ase s, pa allel o dis ibu ed adi ional models a e ecommended. Da ase size is pi o al, in luencing memo y managemen , da a ans e , and communica ion la ency, pa icula ly in MPI-based me hods when using dis ibu ed memo y o capaci y. This issue does no a ec mul i- h eaded me hods (e.g. OpenMP o POSIX Th eads), which ely on a single compu e ’s ha dwa e bu p eclude building CPU clus e s o enhanced pe o mance. To op imise mul i- h eaded algo i hm pe o mance, i is ecommended o limi pa allel ask complexi y and comp ess da ase s o con ol p ocessing and s o age cos s. A.López-Fe nández e al. 1123 Page 34 o 52 Pla o ms like Apache Hadoop o Apache Spa k a e ecommended o handling la ge da ase s equi ing inc eased p ocessing cos s and compu a ional esou ces, acili a ing he de elopmen o biclus e ing solu ions o such da ase s. Howe e , he e exis s a gap be ween he numbe o biclus e ing algo i hms designed o hese pla o ms and hose o o he applica ions aimed a imp o ing compu a ional pe o - mance. Adap ing biclus e ing algo i hms o hese pla o ms o cons uc biclus e s ia da ase pa i ioning poses c i ical challenges such as he numbe o MapReduce jobs, in elligen da ase pa i ioning s a egies, educed I/O ope a ions, and asyn- ch onous communica ion be ween map and educe unc ions. Fu he mo e, hese biclus e ing echniques o en exhibi poo pe o mance on small da ase s due o pla - o m ini ialisa ion ime, signi ican ly educing algo i hm e icacy. Some compa a- i e s udies ha e shown ha biclus e ing algo i hms de eloped using Apache Spa k o e supe io compu a ional pe o mance compa ed o Apache Hadoop, a ibu ed o educed I/O ope a ion cos s and as e access speeds. Recen ly, GPU de ices and he CUDA pla o m ha e become popula o accel- e a ing biclus e ing algo i hms, demons a ing op imal compu a ional pe o mance. Howe e , no all GPU-based biclus e ing algo i hms can handle huge da ase s e ec- i ely, which poses a challenge in de eloping algo i hms o such olumes. Maxim- ising GPU powe u ilisa ion h ough esou ce planning is c ucial, alongside mini- mising da a ans e s be ween CPU and GPU o mi iga e bandwid h limi a ions. While sha ed memo y o e s speed, i s use may lead o memo y o e lows in p o- cessing massi e da a olumes, equi ing ca e ul selec ion o i s usage. Mul i-GPU a chi ec u es can enhance pe o mance, bu wo kload dis ibu ion and synch onisa- ion among GPU de ices a e essen ial. This dis ibu ion ypically employs pa allel o dis ibu ed me hods like OpenMP, POSIX Th eads, o MPI, while asynch onous communica ion ac oss GPU de ices has been shown o imp o e compu a ional pe - o mance by omi ing synch onisa ion. 7.3 Challenges in alida ion Biclus e ing algo i hms o en gene a e a as numbe o esul s o e la ge da ase s, which equi e subsequen alida ion [17, 29]. Typically, alida ion in ol es s a is i- cal o biological knowledge-based me hods, o en sou ced om public da abases, o asce ain he biological ele ance o he gene a ed ou comes [127]. Recen ad ance- men s in biclus e alida ion me hods and ools ha e p ima ily ocused on imp o - ing esul accu acy and de eloping use - iendly in e aces. Howe e , he e is a no able absence o me hodologies and ools capable o acili a ing e icien compa - a i e analysis o biclus e alida ion app oaches. Fo ins ance, EBIC, a mul i-GPU e olu iona y biclus e ing me hod, limi ed i s ou comes o 100 and alida ed hem using he sequen ial so wa e RGOS a s [160], which ope a es solely wi h gene lis s. The e is a p essing need o de elop alida ion echniques o la ge biclus e da ase s. These echniques would enable esea che s o alida e biclus e s de i ed Biclus e ing inbioin o ma ics using big da a andHigh… Page 35 o 52 1123 om da ase s managed by scien i ic ini ia i es o gene a ed om nex -gene a ion sequencing. Addi ionally, i has been demons a ed ha i is no possible o d aw in e es ing biological knowledge om he huge numbe o biclus e s gene a ed by Big Da a-adap ed biclus e ing algo i hms. 7.4 Visualisa ion andin e p e abili y issues A e biclus e ing esul s a e alida ed, esea che s mus ocus on isualising and in e p e ing hem in biological and biomedical con ex s o de i e eliable conclu- sions. Se e al isualisa ion echniques aid in in e p e ing biclus e ing algo i hm ou comes. BicO e lappe [185] add esses he challenge o isualising biclus e o e lap by using in e sec ing hulls, enabling he ex ac ion o biological insigh s h ough in e ac i e isualisa ion. BiVisu [186] employs Pa allel Coo dina e (PC) plo s and objec i e me ics like MSR [14] and A e age Co ela ion Value (ACV) [187] o assess biclus e homogenei y. BiDo s [188] explo es new isual and in e ac- i e me hods o in es iga ing weigh ed biclus e s ac oss domains. O he conside a- ions in biclus e isualisa ion include he limi a ions o hea maps and PC plo s in demons a ing biological ele ance [189]. The au ho s ad oca e o no el isualisa- ion echniques and p opose s anda ds o biclus e ing algo i hm ou pu s o acili a e isualisa ion ool in eg a ion. Addi ionally, ne wo k-based isualisa ion app oaches simpli y he in e p e a ion o biclus e isualisa ion ne wo ks [159, 190]. These isualisa ion ools can help assess he cohe ence o biclus e s o med by biclus e ing echniques, indica ing hei po en ial signi icance. Howe e , i is s ill possible ha some biclus e s hold domain- ele an knowledge [146]. Ex e nal sou ces o domain expe e alua ion may be needed o con i m he au hen ici y o hese biclus e s. Despi e e o s o s anda dise biclus e ing algo i hm ou pu s and u ilise isualisa ion and en ichmen analysis ools o be e biological in e p e a- ion, he shee olume o biclus e s gene a ed in a Big Da a se ing can lead o a loss o unc ionali y. Consequen ly, he e is a need o isualisa ion ools ailo ed o enhance biological in e p e a ion and manage la ge biclus e da ase s. 8 Conclusions The g owing olume o biological and biomedical da a necessi a es he adap a ion o adi ional biclus e ing me hods o e ec i ely manage la ge-scale da ase s. To mee his challenge, ecen ad ancemen s ha e ocused on employing High Pe o mance Compu ing (HPC) and Big Da a amewo ks—such as Apache Spa k, GPU accel- e a ion, and dis ibu ed/pa allel a chi ec u es— o imp o e compu a ional e iciency. Al hough hese me hodologies o e subs an ial bene i s, hey also in oduce chal- lenges ela ed o communica ion o e head, memo y managemen , and ini ialisa ion A.López-Fe nández e al. 1123 Page 36 o 52 la ency, pa icula ly in MPI-based and MapReduce amewo ks. GPU-based solu- ions, while powe ul, equi e ca e ul esou ce alloca ion, CUDA g id op imisa- ion, and e icien da a ans e be ween RAM and GPU memo y o achie e peak pe o mance. Beyond compu a ional conside a ions, alida ion is essen ial o con i ming he eliabili y and biological signi icance o biclus e ing ou comes. Con empo- a y alida ion s a egies—based on s a is ical consis ency o gene en ichmen analysis—s uggle o scale wi h he as numbe o biclus e s p oduced by mod- e n algo i hms. Consequen ly, many ools limi hei assessmen s o a subse o esul s, highligh ing he need o mo e scalable and au oma ed alida ion ech- niques ha le e age HPC in as uc u e. This s udy has iden i ied mul iple a enues o imp o emen . Fu u e esea ch should ocus on de eloping biclus e ing algo i hms ha a e inhe en ly scal- able and op imised o dis ibu ed and he e ogeneous compu ing en i onmen s. Imp o ing he in e p e abili y o biclus e ing esul s is also essen ial o acili a e adop ion by biomedical esea che s, which may in ol e he use o in e p e able models o isualisa ion ools ha cla i y he biological ele ance o he de ec ed pa e ns. S anda dised, open benchma king amewo ks ha ely on eal-wo ld da ase s a e also needed o suppo ep oducible and compa a i e assessmen s. One p omising app oach in ol es in eg a ing p io biological knowledge— such as unc ional anno a ions o egula o y ne wo ks—in o he biclus e ing p ocess o guide he disco e y o biologically meaning ul s uc u es. The ield can also bene i om he inco po a ion o ad anced machine lea ning me hods, including deep lea ning, g aph-based models, and eme ging o ms o a i icial in elligence such as ein o cemen lea ning and ans o me s. These echnolo- gies a e expec ed o play a cen al ole in he nex gene a ion o biclus e ing me hods, enabling mo e adap i e, con ex -awa e, and au oma ed disco e y o complex pa e ns in omics da a. Expanding he applica ion o biclus e ing o mul i-omics in eg a ion and ime se ies analysis could p o ide mo e comp ehensi e insigh s in o dynamic bio- logical sys ems. Fu he mo e, u u e algo i hms should enhance obus ness o noise h ough he use o obus s a is ical echniques and p ep ocessing s a - egies. S eng hening collabo a ion wi h expe imen al biologis s is also key o alida ing compu a ional indings in i o o in i o, he eby suppo ing hei ansla ional ele ance. Looking ahead, an icipa ed mains eam ends include he in eg a ion o biclus e ing in o au oma ed machine lea ning (Au oML) pipelines, enabling non-expe use s o conduc explo a o y analyses in biomedical con ex s. Addi- ionally, he deploymen o biclus e ing wo k lows wi hin cloud-based HPC en i onmen s will enhance scalabili y and accessibili y. Ano he impo an di ec ion is he explo a ion o ede a ed lea ning amewo ks o enable secu e Biclus e ing inbioin o ma ics using big da a andHigh… Page 37 o 52 1123 and p i acy-p ese ing biclus e ing ac oss dis ibu ed biomedical da ase s—a c ucial conside a ion in mul i-ins i u ional and clinical se ings. By add essing hese limi a ions and oppo uni ies, nex -gene a ion biclus e - ing echniques can e ol e in o powe ul, in e p e able, and scalable ools o analysing high-dimensional biological da a. The ising adop ion o deep lea n- ing and AI-d i en me hodologies in bioin o ma ics ep esen s a p omising di ec- ion. In es iga ing how such models can be e ec i ely adap ed wi hin HPC and Big Da a ecosys ems—pa icula ly in alignmen wi h he analy ical s a egies discussed in his pape —cons i u es a signi ican a ea o u u e esea ch. Appendix A: De ailed ables See Tables2, 3, 4, 5, 6, 7, 8 and 9 All he ables ha a e p esen ed below a e e e enced in he main ex . Table 2 A summa y o he mos signi ican biclus e ing e iews Yea Au ho s Applica ion Me hods Valida ion 2004 Madei a e al. [19] Gene exp ession X 2005 Tanay e al. [23] Gene exp ession X 2007 San ama ía e al. [24] Gene exp ession X 2008 Busygin e al. [20] Biomedicine and ex mining X 2010 Ras ega e al. [191] Gene exp ession X 2010 Ve ma e al. [192] Gene exp ession X 2013 E en e al. [25] Gene exp ession X 2013 O zechowski. [128] Gene exp ession X 2014 Oghabian e al. [193] Gene exp ession X 2014 Ho a e al. [194] Gene exp ession X 2015 Pon es e al. [195] Gene exp ession X 2015 Pon es e al. [21] Gene exp ession X 2015 Mouni . [196] Gene exp ession X 2016 Biswal e al. [197] Gene exp ession X 2016 Ani ha e al. [198] Gene exp ession X 2017 Padilha e al. [26] Gene exp ession X X 2018 Biswal e al. [199] Gene exp ession X 2018 Aouabed e al. [200] Biomedical X 2019 Xie e al. [6] Biological and biomedical X 2021 Nicholls e al. [27] Gene exp ession X 2021 Sozdinle , [189] Gene exp ession X 2022 No onha e al. [146] Da a mining and gene exp ession X 2022 José-Ga cía e al. [22] Gene exp ession X 2024 Cas anho e al. [28] Biological and biomedical X X A.López-Fe nández e al. 1123 Page 38 o 52 Table 3 Reposi o ies whe e he sou ce codes o he HPC biclus e ing algo i hms a e s o ed Yea Me hod Sou ce code URL 2007 RoBA [74] – 2008 DisCo [107] – 2008 P-Biclus e [77] – 2009 A.Nisa e al [80] – 2012 FLOC [113] – 2013 GBC [17] – 2014 GBC [117] – 2014 MCC [106] – 2014 BiTM-MR [100]h ps:// gi hub. com/ Tugdu alSa azin/ spa k- clus e ing 2014 Cloudnm [101]h p:// admis. udan. edu. cn/ p oje c s/ Cloud NMF. h ml 2015 NMF [118]h ps:// gi hub. com/ bioin o- cnb/ bionm - gpu 2015 Bha naga e al. [108] – 2015 MFCM [83] – 2016 MR-GABiT [104] – 2017 CCS [120]h ps:// gi hub. com/ abha a3/ Condi ion- depen den - Co e la ion- Subg oups- CCS 2017 PBD-SPEA2 [96] - 2018 EBIC [121]h ps:// gi hub. com/ Epis asisL ab/ ebic 2018 Runibic [85]h ps:// www. bioco nduc o . o g/ packa ges/ elea se/ bioc/ h ml/ unib ic. h ml 2018 Pa BiBi [88]h ps:// sou c e o ge. ne / p oje c s/ pa bi bi / 2019 CuBiBi [122]h ps:// sou c e o ge. ne / p oje c s/ cubib i 2019 SP-PLSS [4] – 2021 ScalaPa BiBi [91]h ps:// gi hub. com/ agu ela/ Scala Pa Bi Bi 2021 COBRAC [92]h ps:// gi hub. com/ haidyi/ c xbi clus 2021 gBiBi [90]h ps:// gi hub. com/ au el iol d ez/ gbibi Biclus e ing inbioin o ma ics using big da a andHigh… Page 39 o 52 1123 Table 4 Main ea u es o biclus e ing algo i hms accele a ed by adi ional pa allel and dis ibu ed mod- els Da a compa ibili y Yea Me hod Da ase s Compa a i e Noise Da a Bina y Disc e e Con inuous 2007 RoBA [74] 2 0 No Mic oa ays X 2008 P-Biclus e [77] 1 1 No Mic oa ays X X 2009 Nisa e al [80] 0 0 Yes Mic oa ays X 2015 MFCM [83] 1 2 No Mic oa ays X 2017 PBD-SPEA2 [96] 2 1 Yes Mic oa ays X 2018 Runibic [85] 1 1 Yes RNA-Seq X X 2018 Pa BiBi [88] 2 1 No Mic oa ays X 2021 ScalaPa BiBi [91] 0 2 No Mic oa ays X 2021 COBRAC [92] 1 0 Yes Mic oa ays X 2022 ARBic [94] 2 1 No Mic oa ays X 2023 EnsemBic [95] 1 1 Yes Mic oa ays & RNA-Seq X Table 5 Compu a ional ea u es o biclus e ing algo i hms based on adi ional pa allel and dis ibu ed models Yea Me hod Language Accele a ion Op imisa ion Com- munica- ions 2007 RoBA [74] MATLAB MPI No Yes 2008 P-Biclus e [77] ANSI C MPI No No 2009 A.Nisa e al [80] C/C++ MPI Yes Yes 2015 MFCM [83] MATLAB MPI No No 2017 PBD-SPEA2 [96] - - - - 2018 Runibic [85] C/C++ OpenMP No No 2018 Pa BiBi [88] C/C++ MPI/POSIX No Yes 2021 ScalaPa BiBi [91] C/C++ MPI/POSIX Yes Yes 2021 COBRAC [92] C/C++ OpenMP Yes No 2022 ARBic [94] C/C++ OpenMP No No 2023 EnsemBic [95] R - - - A.López-Fe nández e al. 1123 Page 40 o 52 Table 6 Main ea u es o biclus e ing algo i hms suppo ed by MapReduce pla o ms Da a compa ibili y Yea Me hod Pla o m Da ase s Compa a i e Noise Da a Bina y Disc e e Con inuous 2008 DisCo [107] Hadoop 1 0 No - X 2014 MCC [106] - 1 1 No Mic oa ays X 2014 BiTM-MR [100] Spa k 0 0 No - X 2014 Cloudnm [101] Hadoop 2 0 No PPI X 2015 Bha naga e al. [108] Hadoop 1 2 No - X 2016 MR-GABiT [104] MATLAB 1 0 No Mic oa ays X X 2019 SP-PLSS [4] Spa k 2 2 No – X X Biclus e ing inbioin o ma ics using big da a andHigh… Page 41 o 52 1123 Table 7 Compu a ional ea u es o biclus e ing algo i hms based on MapReduce pla o ms Yea Me hod MR Jobs Pa i ioning I/O In-memo y Asyn- ch o- nous Speedup Scalabili y 2008 DisCo [107] 1 0 0 0 0 1 3 2014 MCC [106] 1 0 0 0 0 0 1 2014 BiTM-MR [100] 1 0 1 1 1 1 3 2014 Cloudnm [101] 1 0 0 0 0 0 1 2015 Bha naga e al. [108] 1 0 0 0 0 0 3 2016 MR-GABiT [104] 0 1 0 0 0 1 1 2019 SP-PLSS [4] 1 1 1 1 0 1 3 Table 8 Main ea u es o biclus e ing algo i hms suppo ed by GPGPU pla o ms Da a compa ibili y Yea Me hod Da ase s Compa a i e Noise Da a Bina y Disc e e Con inuous 2012 FLOC [113] 0 1 No Mic oa ays X X 2013 GBC [17] 1 2 No Mic oa ays X 2014 GBC [117] 1 2 No Mic oa ays X 2015 NMF [118] 1 1 No Mic oa ays X 2017 CCS [120] 2 1 No Mic oa ays X X 2018 EBIC [121] 2 2 Yes Mic oa ays X X X 2019 CuBiBi [122] 0 1 No Mic oa ays X 2021 gBiBi [90] 2 2 No Mic oa ays & RNA-Seq X Table 9 Compu a ional ea u es o biclus e ing algo i hms based on GPGPU pla o ms Yea Me hod mul i-GPU Sha ed mem. Coalescing Alloca ion T ans e Occupancy Sync. 2012 FLOC [113]No No No 0 Global Low No 2013 GBC [17]No Yes Yes 1 Global Low No 2014 GBC [117]No Yes Yes 1 Global High Yes 2015 NMF [118] Yes (MPI) Yes Yes 1 Global High Yes 2017 CCS [120]No Yes No 1 Global Low No 2018 EBIC [121] Yes (OpenMP) Yes Yes 1 Global High Yes 2019 CuBiBi [122] Yes (POSIX) Yes Yes 1 Global Low No 2021 gBiBi [90] Yes (POSIX) No Yes 1 Global High No A.López-Fe nández e al. 1123 Page 48 o 52 115. Zhao H, Liew AW-C, Xie X, Yan H (2008) A new geome ic biclus e ing algo i hm based on he Hough ans o m o analysis o la ge-scale mic oa ay da a. J Theo Biol 251(2):264–274. h ps:// doi. o g/ 10. 1016/j. j bi. 2007. 11. 030 116. Gandha e S, Ka hikeyan B (2019) Su ey on pga a chi ec u e and ecen applica ions. In: 2019 In e na ional Con e ence on Vision Towa ds Eme ging T ends in Communica ion and Ne wo king (ViTECoN), pp. 1–4. h ps:// doi. o g/ 10. 1109/ ViTEC oN. 2019. 88995 50 117. Liu B, Xin Y, Cheung RC, Yan H (2014) GPU-based biclus e ing o mic oa ay da a analysis in neu ocompu ing. Neu ocompu ing 134:239–246. h ps:// doi. o g/ 10. 1016/j. neucom. 2013. 06. 049 118. Mejía-Roa E, Tabas-Mad id D, Se oain J, Ga cía C, Ti ado F, Pascual-Mon ano A (2015) NMF- mGPU: non-nega i e ma ix ac o iza ion on mul i-GPU sys ems. BMC Bioin o ma 16(1):1– 12. h ps:// doi. o g/ 10. 1186/ s12859- 015- 0485-4 119. Mejía-Roa E, Ga cía C, Gómez JI, P ie o M, Ti ado F, Nogales R, Pascual-Mon ano A (2011) Biclus e ing and classi ica ion analysis in gene exp ession using nonnega i e ma ix ac o i- za ion on mul i-GPU sys ems. In: 2011 11 h In e na ional Con e ence on In elligen Sys ems Design and Applica ions, pp. 882–887. h ps:// doi. o g/ 10. 1109/ ISDA. 2011. 61217 69 120. Bha acha ya A, Cui Y (2017) A GPU-accele a ed algo i hm o biclus e ing analysis and de ec ion o condi ion-dependen coexp ession ne wo k modules. Sci Rep 7(1):1–9. h ps:// doi. o g/ 10. 1038/ s41598- 017- 04070-4 121. O zechowski P, Sippe M, Huang X, Moo e JH (2018) EBIC: an e olu iona y-based pa allel biclus e ing algo i hm o pa e n disco e y. Bioin o ma ics 34(21):3719–3726. h ps:// doi. o g/ 10. 1093/ bioin o ma ics/ b y401 122. González-Domínguez J, Expósi o RR (2019) Accele a ing bina y biclus e ing on pla o ms wi h CUDA-enabled GPUs. In Sci 496:317–325. h ps:// doi. o g/ 10. 1016/j. ins. 2018. 05. 025 123. López-Fe nández A, Gómez-Vela FA, Gonzalez-Dominguez J, Bida e-Di aka acha i P (2024) bioscience: a new py hon science lib a y o high-pe o mance compu ing bioin o ma ics ana- ly ics. So wa eX 26:101666. h ps:// doi. o g/ 10. 1016/j. so x. 2024. 101666 124. A dlie KG, Deluca DS, Seg è AV, Sulli an TJ, Young TR, Gel and ET, T owb idge CA, Malle JB, Tukiainen T, Conso ium G e al (2015) The geno ype- issue exp ession (GTEx) pilo analy- sis: mul i issue gene egula ion in humans. Science 348(6235):648–660. h ps:// doi. o g/ 10. 1126/ scien ce. 12621 10 125. López-Fe nández A, Gómez-Vela FA, Gonzalez-Dominguez J, Bida e-Di aka acha i P (2024) bioscience: a new py hon science lib a y o high-pe o mance compu ing bioin o ma ics ana- ly ics. So wa eX 26:101666. h ps:// doi. o g/ 10. 1016/j. so x. 2024. 101666 126. Ke G, Ruskin HJ, C ane M, Doolan P (2008) Techniques o clus e ing gene exp ession da a. Compu Biol Med 38(3):283–293. h ps:// doi. o g/ 10. 1016/j. compb iomed. 2007. 11. 001 127. Sabe HB, Elloumi M (2015) A new s udy on biclus e ing ools, biclus e s alida ion and e alu- a ion unc ions. In J Compu Sci Eng Su 6(1):1. h ps:// doi. o g/ 10. 5121/ ijcses. 2015. 6101 128. O zechowski P (2013) P oximi y measu es and esul s alida ion in biclus e ing–a su ey. In: In e na ional Con e ence on A i icial In elligence and So Compu ing, pp. 206–217. h ps:// doi. o g/ 10. 1007/ 978-3- 642- 38610-7_ 20 129. Choi S-S, Cha S-H, Tappe CC (2010) A su ey o bina y simila i y and dis ance measu es. J Sys , Cybe n In o ma 8(1):43–48 130. Wang L, Zhang H, Chang H-W, Qin Q-M, Zhang B-R, Li X-Q, Zhao T-H, Zhang T-Y (2021) GAEBic: a no el biclus e ing analysis me hod o miRNA- a ge ed gene da a based on g aph au oencode . J Compu Sci Technol 36(2):299–309. h ps:// doi. o g/ 10. 1007/ s11390- 021- 0804-3 131. Se ano-Rubio AA, Mo ales-Luna GB, Meneses-Vi e os A (2021) Gene exp ession analysis h ough pa allel non-nega i e ma ix ac o iza ion. Compu a ion 9(10):106. h ps:// doi. o g/ 10. 3390/ compu a io n9100 106 132. Beli se E, Nu ushe N (2018) Local in e ence by penaliza ion me hod o biclus e ing model. Ma h Me hods S a is 27(3):163–183. h ps:// doi. o g/ 10. 3103/ S1066 53071 80300 18 133. Wang H, Wang W, Yang J, Yu PS (2002) Clus e ing by pa e n simila i y in la ge da a se s. In: P oceedings o he 2002 ACM SIGMOD In e na ional Con e ence on Managemen o Da a, pp. 394–405. h ps:// doi. o g/ 10. 1145/ 564691. 564737 134. Xiao F, Chen L, Sha C, Sun L, Wang R, Liu AX, Ahmed F (2018) Noise ole an localiza ion o senso ne wo ks. IEEE/ACM T ans Ne w 26(4):1701–1714. h ps:// doi. o g/ 10. 1109/ TNET. 2018. 28527 54 Biclus e ing inbioin o ma ics using big da a andHigh… Page 49 o 52 1123 135. Mao KZ, Tang W (2010) Recu si e mahalanobis sepa abili y measu e o gene subse selec ion. IEEE/ACM T ans Compu Biol Bioin 8(1):266–272. h ps:// doi. o g/ 10. 1109/ TCBB. 2010. 43 136. Naja N, Abdulazeez AM (2017) Gene clus e ing wi h pa i ion a ound mediods algo i hm based on weigh ed and no malized mahalanobis dis ance. In: 2017 In e na ional Con e ence on In elli- gen In o ma ics and Biomedical Sciences (ICIIBMS), pp. 140–145. h ps:// doi. o g/ 10. 1109/ ICIIB MS. 2017. 82797 07 137. Yip KY, Cheung DW, Ng MK (2004) Ha p: a p ac ical p ojec ed clus e ing algo i hm. IEEE T ans Knowl Da a Eng 16(11):1387–1397. h ps:// doi. o g/ 10. 1109/ TKDE. 2004. 74 138. Aguila -Ruiz JS (2005) Shi ing and scaling pa e ns om gene exp ession da a. Bioin o ma ics 21(20):3840–3845. h ps:// doi. o g/ 10. 1093/ bioin o ma ics/ b i641 139. Bozdağ D, Kuma AS, Ca alyu ek UV (2010) Compa a i e analysis o biclus e ing algo i hms. In: P oceedings o he Fi s ACM In e na ional Con e ence on Bioin o ma ics and Compu a ional Biology, pp. 265–274. h ps:// doi. o g/ 10. 1145/ 18547 76. 18548 14 140. Chen PY, Smi hson M, Popo ich PM, Y C, e al.: Co ela ion: Pa ame ic and Nonpa ame ic Measu es ol. 139, (2002). Sage 141. P iness I, Maimon O, Ben-Gal I (2007) E alua ion o gene-exp ession clus e ing ia mu ual in o - ma ion dis ance measu e. BMC Bioin o ma 8(1):1–12. h ps:// doi. o g/ 10. 1186/ 1471- 2105-8- 111 142. Li C, Tang Z, Zhang W, Ye Z, Liu F (2021) GEPIA2021: in eg a ing mul iple decon olu ion-based analysis in o GEPIA. Nucleic Acids Res 49(W1):242–246. h ps:// doi. o g/ 10. 1093/ na / gkab4 18 143. Roy S, Bha acha yya D, Kali a JK (2012) De e minis ic app oach o biclus e ing o co- egula ed genes om gene exp ession da a. In: Ad ances in Knowledge-Based and In elligen In o ma ion and Enginee ing Sys ems, pp. 490–499. h ps:// doi. o g/ 10. 3233/ 978-1- 61499- 105-2- 490 144. Ayadi W, Elloumi M, Hao J-K (2012) Pa e n-d i en neighbo hood sea ch o biclus e - ing o mic oa ay da a. In: BMC Bioin o ma ics, ol. 13, pp. 1–11. h ps:// doi. o g/ 10. 1186/ 1471- 2105- 13- S7- S11 145. D’U so P (2015) Fuzzy clus e ing. Handbook o clus e analysis, 545–574 146. No onha MD, Hen iques R, Madei a SC, Zá a e LE (2022) Impac o me ics on biclus e ing solu- ion and quali y: a e iew. Pa e n Recogni . h ps:// doi. o g/ 10. 1016/j. pa cog. 2022. 108612 147. Conso ium GO (2019) The gene on ology esou ce: 20 yea s and s ill Going s ong. Nucleic Acids Res 47(D1):330–338. h ps:// doi. o g/ 10. 1093/ na / gky10 55 148. Kanehisa M, Go o S (2000) KEGG: Kyo o encyclopedia o genes and genomes. Nucleic Acids Res 28(1):27–30. h ps:// doi. o g/ 10. 1093/ na / 28.1. 27 149. Zou D, Ma L, Yu J, Zhang Z (2015) Biological da abases o human esea ch. Genom, P o eomics & Bioin o ma 13(1):55–63. h ps:// doi. o g/ 10. 1016/j. gpb. 2015. 01. 006 150. Akoglu H (2018) Use ’s guide o co ela ion coe icien s. Tu kish J Eme g Med 18(3):91–93. h ps:// doi. o g/ 10. 1016/j. jem. 2018. 08. 001 151. Raud e e U, Kolbe g L, Kuzmin I, A ak T, Adle P, Pe e son H, Vilo J (2019) G: P o ile : a web se e o unc ional en ichmen analysis and con e sions o gene lis s (2019 upda e). Nucleic Acids Res 47(W1):191–198. h ps:// doi. o g/ 10. 1093/ na / gkz369 152. Kulesho MV, Jones MR, Rouilla d AD, Fe nandez NF, Duan Q, Wang Z, Kople S, Jenkins SL, Jagodnik KM, Lachmann A e al (2016) En ich : a comp ehensi e gene se en ichmen analysis web se e 2016 upda e. Nucleic Acids Res 44(W1):90–97. h ps:// doi. o g/ 10. 1093/ na / gkw377 153. Fan J, Fan D, Slowikowski K, Gehlenbo g N, Kha chenko P (2017) UBiT2: a clien -side web- applica ion o gene exp ession da a analysis. bioRxi , 118992 h ps:// doi. o g/ 10. 1101/ 118992 154. Sun L, Zhu Y, Mahmood A, Tudo CO, Ren J, Vijay-Shanke K, Chen J, Schmid CJ (2017) Web- GIVI: a web-based gene en ichmen analysis and isualiza ion ool. BMC Bioin o ma 18(1):1–10. h ps:// doi. o g/ 10. 1186/ s12859- 017- 1664-2 155. Liao Y, Wang J, Jaehnig EJ, Shi Z, Zhang B (2019) WebGes al 2019: gene se analysis oolki wi h e amped UIs and APIs. Nucleic Acids Res 47(W1):199–205. h ps:// doi. o g/ 10. 1093/ na / gkz401 156. Tipney H, Hun e L (2010) An in oduc ion o e ec i e use o en ichmen analysis so wa e. Hum Genom 4(3):1–5. h ps:// doi. o g/ 10. 1186/ 1479- 7364-4- 3- 202 157. Hung J-H, Yang T-H, Hu Z, Weng Z, DeLisi C (2012) Gene se en ichmen analysis: pe o mance e alua ion and usage guidelines. B ie Bioin o m 13(3):281–291. h ps:// doi. o g/ 10. 1093/ bib/ bb 049 158. Hoadley KA, Yau C, Wol DM, Che niack AD, Tambo e o D, Ng S, Leise son MD, Niu B, McLel- lan MD, Uzunangelo V e al (2014) Mul ipla o m analysis o 12 cance ypes e eals molecula classi ica ion wi hin and ac oss issues o o igin. Cell 158(4):929–944. h ps:// doi. o g/ 10. 1016/j. cell. 2014. 06. 049 A.López-Fe nández e al. 1123 Page 50 o 52 159. Lopez-Fe nandez A, Rod iguez-Baena D, Gomez-Vela F, Diaz-Diaz N (2018) BIGO: A web appli- ca ion o analyse gene en ichmen analysis esul s. Compu Biol Chem 76:169–178. h ps:// doi. o g/ 10. 1016/j. compb iolch em. 2018. 06. 006 160. Falcon S, Gen leman R (2007) Using GOs a s o es gene lis s o GO e m associa ion. Bioin o - ma ics 23(2):257–258. h ps:// doi. o g/ 10. 1093/ bioin o ma ics/ b l567 161. Gomez-Pulido JA, Ce ada-Ba ios JL, T inidad-Amado S, Lanza-Gu ie ez JM, Fe nandez-Diaz RA, C aw o d B, So o R (2016) Fine-g ained pa alleliza ion o i ness unc ions in bioin o ma ics op imiza ion p oblems: gene selec ion o cance classi ica ion and biclus e ing o gene exp ession da a. BMC Bioin o ma 17(1):1–13. h ps:// doi. o g/ 10. 1186/ s12859- 016- 1200-9 162. Bou os A, Be z V (2021) FPGA a chi ec u e: p inciples and p og ession. IEEE Ci cui s Sys Mag 21(2):4–29. h ps:// doi. o g/ 10. 1109/ MCAS. 2021. 30716 07 163. López-Fe nández A, Rod íguez-Baena DS, Gómez-Vela F (2020) gMSR: a mul i-GPU algo i hm o accele a e a massi e alida ion o biclus e s. Elec onics 9(11):1782. h ps:// doi. o g/ 10. 3390/ elec onic s9111 782 164. G eene CS, Tan J, Ung M, Moo e JH, Cheng C (2014) Big da a bioin o ma ics. J Cell Physiol 229(12):1896–1900. h ps:// doi. o g/ 10. 1002/ jcp. 24662 165. DiMaggio PA, McAllis e SR, Floudas CA, Feng X-J, Rabinowi z JD, Rabi z HA (2008) Biclus e - ing ia op imal e-o de ing o da a ma ices in sys ems biology: igo ous me hods and compa a i e s udies. BMC Bioin o ma 9(1):1–16. h ps:// doi. o g/ 10. 1186/ 1471- 2105-9- 458 166. Wo dswo h S, Doble B, Payne K, Buchanan J, Ma shall DA, McCabe C, Regie DA (2018) Using ‘big da a’ in he cos -e ec i eness analysis o nex -gene a ion sequencing echnologies: challenges and po en ial solu ions. Value Heal h 21(9):1048–1053. h ps:// doi. o g/ 10. 1016/j. j al. 2018. 06. 016 167. Scha z MC, Langmead B, Salzbe g SL (2010) Cloud compu ing and he DNA da a ace. Na Bio- echnol 28(7):691–693. h ps:// doi. o g/ 10. 1038/ nb 07 10- 691 168. Luo Y, Hi z BC, Gabdank I, Hil on JA, Kagda MS, Lam B, Mye s Z, Sud P, Jou J, Lin K e al (2020) New de elopmen s on he encyclopedia o DNA elemen s (ENCODE) da a po al. Nucleic Acids Res 48(D1):882–889. h ps:// doi. o g/ 10. 1093/ na / gkz10 62 169. Can elli G, Ba eman A, B ooksbank C, Pe o AI, Malik-She i RS, Ide-Smi h M, He mjakob H, Flicek P, Apweile R, Bi ney E e al (2022) The Eu opean Bioin o ma ics Ins i u e (EMBL-EBI) in 2021. Nucleic Acids Res 50(D1):11–19. h ps:// doi. o g/ 10. 1093/ na / gkab1 127 170. Bao R, Huang L, And ade J, Tan W, Kibbe WA, Jiang H, Feng G (2014) Re iew o cu en me h- ods, applica ions, and da a managemen o he bioin o ma ics analysis o whole exome sequenc- ing. Cance In o ma 13:13779. h ps:// doi. o g/ 10. 4137/ CIN. S13779 171. Pabinge S, Dande A, Fische M, Snajde R, Spe k M, E emo a M, K abichle B, Speiche MR, Zschocke J, T ajanoski Z (2014) A su ey o ools o a ian analysis o nex -gene a ion genome sequencing da a. B ie Bioin o m 15(2):256–278. h ps:// doi. o g/ 10. 1093/ bib/ bbs086 172. O’D iscoll A, Daugelai e J, Slea o RD (2013) ‘Big da a’, Hadoop and cloud compu ing in genom- ics. J Biomed In o m 46(5):774–781. h ps:// doi. o g/ 10. 1016/j. jbi. 2013. 07. 001 173. Luo J, Wu M, Gopukuma D, Zhao Y (2016) Big da a applica ion in biomedical esea ch and heal h ca e: a li e a u e e iew. Biomed In o ma Insigh s 8:31559. h ps:// doi. o g/ 10. 4137/ BII. S31559 174. Smow on C, Balla A, An oniades D, Mille C, Pallis G, Dikaiakos MD, Xing W (2017) A cos - e ec i e app oach o imp o ing pe o mance o big genomic da a analyses in clouds. Fu u Gene Compu Sys 67:368–381. h ps:// doi. o g/ 10. 1016/j. u u e. 2015. 11. 011 175. Schad EE, Linde man MD, So enson J, Lee L, Nolan GP (2011) Cloud and he e ogeneous compu ing solu ions exis oday o he eme ging big da a p oblems in biology. Na Re Gene 12(3):224–224. h ps:// doi. o g/ 10. 1038/ n g28 57- c2 176. G ossman RL, Whi e KP (2012) A ision o a biomedical cloud. J In e n Med 271(2):122–130. h ps:// doi. o g/ 10. 1111/j. 1365- 2796. 2011. 02491.x 177. Kuma S (2021) T ends and ad ancemen s in genome da a comp ession and p ocessing algo i hms. So Compu : Theo Appl. h ps:// doi. o g/ 10. 1007/ 978- 981- 16- 1696-9_ 15 178. Kaushik P, Rao AM, Singh DP, Vashish S, Gup a S (2021) Cloud Compu ing and Compa ison based on Se ice and Pe o mance be ween Amazon AWS, Mic oso Azu e, and Google Cloud. In: 2021 In e na ional Con e ence on Technological Ad ancemen s and Inno a ions (ICTAI), pp. 268–273. h ps:// doi. o g/ 10. 1109/ ICTAI 53825. 2021. 96734 25 179. Jain S, Saxena A, Hesa u S, Bhadhadha a K, Bha i N, Kasibha la SM, Sona ane U, Joshi R (2021) GenoVaul : a cloud based genomics eposi o y. BioDa a Mining 14(1):1–10. h ps:// doi. o g/ 10. 1186/ s13040- 021- 00268-5 Biclus e ing inbioin o ma ics using big da a andHigh… Page 51 o 52 1123 180. Schumache A, Pi eddu L, Niemenmaa M, Kallio A, Ko pelainen E, Zane i G, Heljanko K (2014) SeqPig: simple and scalable sc ip ing o la ge sequencing da a se s in Hadoop. Bioin o ma ics 30(1):119–120. h ps:// doi. o g/ 10. 1093/ bioin o ma ics/ b 601 181. Abuín JM, Pichel JC, Pena TF, Amigo J (2015) BigBWA: app oaching he Bu ows-Wheele aligne o big da a echnologies. Bioin o ma ics 31(24):4003–4005. h ps:// doi. o g/ 10. 1093/ bioin o ma ics/ b 506 182. Masse oli M, Canakoglu A, Pinoli P, Kai oua A, Gulino A, Ho lo a O, Nanni L, Be nasconi A, Pe na S, S amoulaka ou E e al (2019) P ocessing o big he e ogeneous genomic da ase s o e - ia y analysis o nex gene a ion sequencing da a. Bioin o ma ics 35(5):729–736. h ps:// doi. o g/ 10. 1093/ bioin o ma ics/ b y688 183. Wiewió ka M, Szmu ło A, Kuśmi ek W, Gambin T (2019) SeQuiLa-co : a as and scalable lib a y o dep h o co e age calcula ions. Gigascience 8(8):094. h ps:// doi. o g/ 10. 1093/ gigas cience/ giz094 184. Expósi o RR, Galego-To ei o R, González-Domínguez J (2020) Sequal: big da a ool o pe o m quali y con ol and da a p ep ocessing o la ge NGS da ase s. IEEE Access 8, 146075–146084 h ps:// doi. o g/ 10. 1109/ ACCESS. 2020. 30150 16 185. San ama ía R, The ón R, Quin ales L (2008) BicO e lappe : a ool o biclus e isualiza ion. Bio- in o ma ics 24(9):1212–1213. h ps:// doi. o g/ 10. 1093/ bioin o ma ics/ b n076 186. Cheng K-O, Law N-F, Siu W-C, Lau T (2007) BiVisu: so wa e ool o biclus e de ec ion and isualiza ion. Bioin o ma ics 23(17):2342–2344. h ps:// doi. o g/ 10. 1093/ bioin o ma ics/ b m338 187. Teng L, Chan L-W (2006) Biclus e ing gene exp ession p o iles by al e na ely so ing wi h weigh ed co ela ed coe icien . In: 2006 16 h IEEE Signal P ocessing Socie y Wo kshop on Machine Lea ning o Signal P ocessing, pp. 289–294. h ps:// doi. o g/ 10. 1109/ MLSP. 2006. 275563 188. Zhao J, Sun M, Chen F, Chiu P (2017) Bido s: Visual explo a ion o weigh ed biclus e s. IEEE T ans Visual Compu G aphics 24(1):195–204. h ps:// doi. o g/ 10. 1109/ TVCG. 2017. 27444 58 189. Sozdinle M (2021) A Re iew on Analysis and Visualiza ion Me hods o Biclus e ing. a Xi p ep in 190. Me ico D, G elle D, Bade GD (2009) How o isually in e p e biological da a using ne wo ks. Na Bio echnol 27(10):921–924. h ps:// doi. o g/ 10. 1038/ nb . 1567 191. Ras ega -Moja ad M, Tala ian-Azad S, Minaei-Bidgoli B (2010) A su ey on biological da a anal- ysis by biclus e ing. In: 2010 In e na ional Con e ence on Educa ional and In o ma ion Technol- ogy, ol. 1, pp. 1–100. h ps:// doi. o g/ 10. 1109/ ICEIT. 2010. 56077 92 192. Ve ma NK, Meena S, Bajpai S, Singh A, Nag a e A, Cui Y (2010) A compa ison o biclus e ing algo i hms. In: 2010 In e na ional Con e ence on Sys ems in Medicine and Biology, pp. 90–97. h ps:// doi. o g/ 10. 1109/ ICSMB. 2010. 57353 51 193. Oghabian A, Kilpinen S, Hau aniemi S, Czeizle E (2014) Biclus e ing me hods: biological ele- ance and applica ion in gene exp ession analysis. PLoS ONE 9(3):90801. h ps:// doi. o g/ 10. 1371/ jou n al. pone. 00908 01 194. Ho a D, Campello RJ (2014) Simila i y measu es o compa ing biclus e ings. IEEE/ACM T ans Compu Biol Bioin 11(5):942–954. h ps:// doi. o g/ 10. 1109/ TCBB. 2014. 23250 16 195. Pon es B, Gi ldez R, Aguila -Ruiz JS (2015) Quali y measu es o gene exp ession biclus e s. PLoS ONE 10(3):0115497. h ps:// doi. o g/ 10. 1371/ jou n al. pone. 01154 97 196. Mouni M, Hamdy M (2015) On biclus e ing o gene exp ession da a. In: 2015 IEEE Se en h In e na ional Con e ence on In elligen Compu ing and In o ma ion Sys ems (ICICIS), pp. 641– 648. h ps:// doi. o g/ 10. 1109/ In el CIS. 2015. 73972 90 197. Biswal BS, Mish a P, Mohapa a A, Vipsi a S (2016) A su ey on g eedy based algo i hms o biclus e ing o gene exp ession mic oa ay da a. In: 2016 In e na ional Con e ence on In o ma ion Technology (ICIT), pp. 124–128. h ps:// doi. o g/ 10. 1109/ ICIT. 2016. 036 198. Ani ha S, Chand an C (2016) Re iew on analysis o gene exp ession da a using biclus e ing app oaches. Bon ing In J Da a Mining 6(2):16–23. h ps:// doi. o g/ 10. 9756/ BIJDM. 8135 199. Biswal BS, Mohapa a A, Vipsi a S (2018) A e iew on biclus e ing o gene exp ession mic oa - ay da a: algo i hms, e ec i e measu es and alida ions. In J Da a Min Bioin o m 21(3):230–268. h ps:// doi. o g/ 10. 1504/ IJDMB. 2018. 097683 200. Aouabed H, San ama ia R, Elloumi M (2018) Biclus e ing impac in biomedical sciences ia li - e a u e mining. In J Biomed Da a Min 7(134):2. h ps:// doi. o g/ 10. 4172/ 2090- 4924. 10001 34 Publishe ’s No e Sp inge Na u e emains neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a ilia ions. A.López-Fe nández e al. 1123 Page 52 o 52 Au ho s and A ilia ions Au elioLópez‑Fe nández4· F anciscoA.Gomez‑Vela1· DomingoS.Rod iguez‑Baena1· Fe nandoM.Delgado‑Cha es2· Jo geGonzalez‑Dominguez3 * Au elio López-Fe nández [email p o ec ed] F ancisco A. Gomez-Vela [email p o ec ed] Domingo S. Rod iguez-Baena [email p o ec ed] Fe nando M. Delgado-Cha es e nando.miguel.delgado-cha es@uni-hambu g.de Jo ge Gonzalez-Dominguez jo g[email p o ec ed] 1 In elligen Da a Analysis G oup (DATAi), Uni e sidad Pablo de Ola ide, C a. U e a, km. 1, ES-41013Se ille, Spain 2 Ins i u e o Compu a ional Sys ems Biology, Uni e si y o Hambu g, No kes asse 9, 22607Hambu g, Ge many 3 Compu e A chi ec u e G oup, Uni e sidade da Co uña, Campus de El iña, 15071ACo uña, Spain 4 Dp o. Lenguajes y Sis emas In o má icos, Uni e sidad de Se illa, Se ille, Spain