Smart Microfluidics: a curated dataset of microfluidic liposome formulations with cross-laboratory validation for machine-learning applications
Lavagna, Leonardo; Buttitta, Giorgio
- Publisher
- Zenodo
- Language
- en
Abstract
This dataset documents how different formulation choices and microfluidic operating conditions influence the size and uniformity of liposomes produced using a controlled microfluidic system. The data were generated through a multi-step experimental workflow that included a screening phase, two response-surface optimization phases, and an independent cross-laboratory study. Together, these experiments provide an extensive view of how lipid composition and flow conditions shape the final properties of the liposomes. Liposome formation is sensitive to both the ingredients used (such as lipid ratios) and the production settings (such as flow rates, flow-rate ratios, and buffer choice). Factors like mixing intensity, solvent–aqueous ratios, and chip geometry all influence how lipids self-assemble into vesicles. This dataset enables the systematic study of these relationships across a broad and diverse experimental space. The dataset package is contained in the microfluidics_dataset.zip folder and is composed of 5 subfolders: code: This folder contains Python code for preprocessing steps and examples of data exploration and usage. It includes the following: logs: directory containing logging information about the processing scripts in the current directory. data_gn_adder.py: script that generates the dataset extension data_extensions/formulations_extended_with_gn.csv. data_smote_adder.py: script that generates the dataset extension data_extensions/formulations_extended_with_SMOTE.csv. raw_data_checker.py: script that validates and preprocesses formulations in the raw_data directory. raw_data_slicer.py: script that selects the correct CHIP configuration from the files in the raw_data directory. data: This folder contains the preprocessed raw datasets, specifically: formulations.csv: the main cleaned dataset obtained by preprocessing and merging initial_formulations_raw.xlsx with new_formulations_raw.xlsx. wet_lab_validation.csv: contains independent wet-lab validation formulations obtained by preprocessing wet_lab_validation_raw.xlsx. data_extensions: This folder contains two artificially extended datasets: formulations_extended_with_gn.csv, generated by adding Gaussian noise to formulations.csv. formulations_extended_with_SMOTE.csv, generated by applying SMOTE interpolation to formulations.csv. metadata: This folder contains documentation and supporting information, including: features_bounds.json: a dictionary describing the physical constraints of each feature. features_names_mappings.json: a dictionary containing standardized names for each feature and additional data conventions. features_descriptions.txt: a text file providing brief descriptions of each feature in the raw datasets. raw_data: contains the raw spreadsheets compiled by the operators initial_formulations_raw.xlsx was collected by one operator (operator A) using a specific instrument (equipment A) in a laboratory in Latina (Latium), Italy (laboratory A). new_formulations_raw.xlsx was collected by operator A using an equivalent instrument (equipment B) in a laboratory in Rome, Italy (laboratory B). wet_lab_validation_raw.xlsx was collected by another independent operator (operator B) using equipment A in laboratory A. Experimental variables included in each data record encompass formulation components (ESM, HSPC, CHOL, PEG), microfluidic operating conditions (Total Flow Rate, Flow Rate Ratio, aqueous medium), and microfluidic hardware type (Droplet or Micromixer). Output variables include liposome size, polydispersity index (PDI), and a binary formation indicator. Only the Micromixer produced reliable liposomes; therefore, only Micromixer entries are retained in the cleaned dataset. The preprocessing pipeline applies standardized naming, consistency checks, physical constraint validation, chip filtering, outlier removal, and optional augmentation via Gaussian noise or SMOTE. The final cleaned dataset contains 304 high-quality microfluidic liposome formulations suitable for statistical modelling and machine learning. Further details, including full preprocessing steps, variable definitions, and usage examples, are available in the README file included in this package.
Full text
README.md 2025-12-15 1 / 5 A c ur a t ed da t a s e t o f m ic ro f lu idic l i posom e f ormul a t i ons w i t h c ross - l ab or a tory v a l ida t i on f or m achi n e - l ea rn i n g a ppl ica t i ons T hi s da t a s e t d o c um e nts h ow diffe r e nt f ormul a t i on ch o ice s a n d m ic ro f lu idic op e r a t i n g c on di t i ons i n f lu e n ce t he s i z e a n d un if orm i ty o f l i posom e s pro d u ced us i n g a c ontroll ed m ic ro f lu idic syst e m . T he da t a w e r e ge n e r a t ed t h rou gh a mult i - st e p e xp e r i m e nt a l wor kf low t ha t i n c lu ded a s c r ee n i n g p ha s e , two r e spons e - sur face opt i m i z a t i on p ha s e s , a n d a n i n de p e n de nt c ross - l ab or a tory stu d y . T o ge t he r , t he s e e xp e r i m e nts prov ide a n e xt e ns i v e v ie w o f h ow l i p id c ompos i t i on a n d f low c on di t i ons s ha p e t he fi n a l prop e rt ie s o f t he l i posom e s . L i posom e f orm a t i on i s s e ns i t i v e to b ot h t he i n g r edie nts us ed ( su ch a s l i p id r a t i os ) a n d t he pro d u c t i on s e tt i n g s ( su ch a s f low r a t e s , f low - r a t e r a t i os , a n d b u ffe r ch o ice ). F ac tors l ike m i x i n g i nt e ns i ty , solv e nt – a qu e ous r a t i os , a n d chi p ge om e try a ll i n f lu e n ce h ow l i p id s s e l f - a ss e m b l e i nto v e s ic l e s . T hi s da t a s e t e n ab l e s t he syst e m a t ic stu d y o f t he s e r e l a t i ons hi ps ac ross a b ro ad a n d di v e rs e e xp e r i m e nt a l sp ace . I ntro d u c t i on T he da t a s e t p ackage i s c ompos ed o f 5 su bf ol de rs : c o de c ont ai ns P yt h on c o de f or pr e pro ce ss i n g st e ps a n d e x a mpl e s o f da t a e xplor a t i on a n d us age , i n p a rt ic ul a r logs : di r ec tory c ont ai n i n g lo ggi n g i n f orm a t i on ab out t he pro ce ss i n g s c r i pts i n t he c urr e nt di r ec tory . data_gn_adder.py : s c r i pt to ge n e r a t e t he da t a s e t e xt e ns i on data_extensions/formulations_extended_with_gn.csv . data_smote_adder.py : s c r i pt to ge n e r a t e t he da t a s e t e xt e ns i on data_extensions/formulations_extended_with_SMOTE.csv . raw_data_checker.py : s c r i pt to v a l ida t e a n d pr e pro ce ss f ormul a t i ons i n t he raw_data di r ec tory . raw_data_slicer.py : s c r i pt to s e l ec t t he c orr ec t CHIP c on fig ur a t i on f rom t he fi l e s i n t he raw_data di r ec tory . da t a c ont ai ns t he pr e pro ce ss ed r a w da t a s e ts , i n p a rt ic ul a r formulations.csv i s t he m ai n c l ea n ed da t a s e t o b t ai n ed b y pr e pro ce ss i n g a n d m e r gi n g initial_formulations_raw.xlsx w i t h new_formulations_raw.xlsx . wet_lab_validation.csv c ont ai ns i n de p e n de nt w e t - l ab v a l ida t i on f ormul a t i ons o b t ai n ed b y pr e pro ce ss i n g wet_lab_validation_raw.xlsx . da t a _ e xt e ns i ons c ont ai ns two a rt ificia lly e xt e n ded da t a s e ts o b t ai n ed b y addi n g G a uss ia n no i s e to formulations.csv ( y ie l di n g formulations_extended_with_gn.csv ) a n d b y a pply i n g SMO T E i nt e rpol a t i on ( y ie l di n g formulations_extended_with_SMOTE.csv ). m e t ada t a C ont ai ns d o c um e nt a t i on fi l e s su ch a s features_bounds.json : dic t i on a ry w i t h t he p h ys ica l c onstr ai nts o f each fea tur e . features_names_mappings.json : dic t i on a ry c ont ia n i n g st a n da r di z ed n a m e s f or each fea tur e a n d ot he r da t a c onv e nt i ons .
README.md 2025-12-15 2 / 5 features_descriptions.txt : fi l e w i t h a b r ief de s c r i pt i on o f each fea tur e i n t he r a w da t a s e ts . r a w _ da t a c ont ai ns t he r a w spr ead s hee ts c omp i l ed b y t he op e r a tors initial_formulations_raw.xlsx w a s c oll ec t ed b y on e op e r a tor ( op e r a tor A ) us i n g a sp ecific i nstrum e nt ( e qu i pm e nt A ) i n a l ab or a tory i n L a t i n a ( L a t i um ), I t a ly ( l ab or a tory A ). new_formulations_raw.xlsx w a s c oll ec t ed b y op e r a tor A us i n g a n e qu i v a l e nt i nstrum e nt ( e qu i pm e nt B ) i n a l ab or a tory i n R om e , I t a ly ( l ab or a tory B ). wet_lab_validation_raw.xlsx w a s c oll ec t ed b y a not he r i n de p e n de nt op e r a tor ( op e r a tor B ) us i n g e qu i pm e nt A i n l ab or a tory A . T he stru c tur e o f t he da t a s e t f ol de r i s s h own be low . dataset | |_ code | |_ logs | |_ data_gn_adder.log | |_ data_smote_adder.log | |_ raw_data_checker.log | |_ raw_data_slicer.log | | |_ data_gn_adder.py | |_ data_smote_adder.py | |_ raw_data_checker.py | |_ raw_data_slicer.py | |_ data | |_ formulations.csv | |_ wet_lab_validation.csv | |_ data_extensions | |_ formulations_extended_with_gn.csv | |_ formulations_extended_with_SMOTE.csv | |_ metadata | |_ features_bounds.json |
README.md 2025-12-15 3 / 5 |_ features_names_mappings.json | |_ features_descriptions.txt | |_ raw_data | |_ initial_formulations_raw.xlsx | |_ new_formulations_raw.xlsx | |_ wet_lab_validation_raw.xlsx | |_ README.md R a w da t a s e t de s c r i pt i on E ach row i n t he r a w da t a s e t r e pr e s e nts on e c ompl e t e m ic ro f lu idic pro d u c t i on run . F or e v e ry run , t he da t a s e t i n c lu de s b ot h t he i nput s e tt i n g s c ontroll ed b y t he e xp e r i m e nt e r a n d t he outputs m ea sur ed af t e rw a r d s . T he s e i nputs a n d outputs v a ry ac ross t he f ollow i n g di m e ns i ons : F ormul a t i on v a r iab l e s : c on ce ntr a t i ons o f f our c l i n ica lly r e l e v a nt l i p id s t ha t de t e rm i n e m e m b r a n e c ompos i t i on a n d st abi l i ty , i n p a rt ic ul a r ESM : E gg sp hi n g omy e l i n c on ce ntr a t i on ( m g / m L ) HSPC : H y d ro ge n a t ed soy p h osp ha t id yl ch ol i n e c on ce ntr a t i on ( m g / m L ) CHOL : C h ol e st e rol c on ce ntr a t i on ( m g / m L ) PEG : DSPE - PEG 2000 c on ce ntr a t i on ( m g / m L ) P ro ce ss c on di t i ons t ha t affec t m i x i n g efficie n c y a n d l i p id s e l f - a ss e m b ly i ns ide t he chi p , i n p a rt ic ul a r TFR : t he T ot a l F low R a t e , i n dica t i n g t he sp eed a t w hich t he a qu e ous a n d or ga n ic str ea ms w e r e i n jec t ed ( m L / m i n ). FRR : t he F low R a t e R a t i o , s h ow i n g t he r e l a t i v e proport i on o f a qu e ous to or ga n ic p ha s e s ( a qu e ous : or ga n ic ). AQUEOUS or Aqueous medium : t he b u ffe r e nv i ronm e nt us ed , e n c omp a ss i n g two a qu e ous m edia c ommonly e mploy ed i n l i posom e pr e p a r a t i on ( MQ a n d PBS ). M ic ro f lu idic ha r d w a r e i n c lu di n g t he v a r iab l e CHIP : ei t he r D ropl e t or M ic rom i x e r . N ot e . O nly t he M ic ro mi x e r pro d u ced r e l iab l e l i posom e s , so t he c l ea n ed da t a s e t kee ps only t h os e e ntr ie s . O utputs m ea sur ed v ia D yn a m ic L igh t S ca tt e r i n g f ollow i n g t he proto c ols i n t he a sso cia t ed a rt ic l e SIZE or Size : t he dia m e t e r o f t he l i posom e s ( nm ). PDI : a m ea sur e be tw ee n 0 a n d 1 o f s i z e un if orm i ty ( poly di sp e rs i ty i n de x , w he r e 0 m ea ns un e v e n s i z e di str ib ut i on a n d 1 m ea ns un if orm s i z e ). OUTPUT or Formation : a bi n a ry v iabi l i ty f l ag ( r e mov ed i n t he c l ea n ed da t a s e t ) i n dica t i n g w he t he r w e ll - f orm ed l i posom e s w e r e o b t ai n ed .
README.md 2025-12-15 4 / 5 A ux i l ia ry v a r iab l e s i n c lu di n g a n a lp ha num e r ic ID f or each f ormul a t i on a n d a r efe r e n ce to t he Main Lipid ( ei t he r ESM or HSPC ). T he r a w da t a s e t i n c lu de s t h r ee m ai n e xp e r i m e nt a l b lo ck s : S eed da t a s e t ( ca ll ed initial_formulations_raw , n = 276) ge n e r a t ed us i n g a stru c tur ed D e s ig n - o f - E xp e r i m e nts a ppro ach , c ov e r i n g a w ide r a n ge o f f ormul a t i on a n d f low c on di t i ons . C ross - l ab or a tory e xt e ns i on da t a s e t ( ca ll ed new_formulations_raw , n = 58) c ons i st i n g o f i n de p e n de ntly pro d u ced f ormul a t i ons i n a s ec on d l ab us i n g t he s a m e e qu i pm e nt a n d proto c ols . T hi s port i on e v a lu a t e s r e pro d u cibi l i ty , op e r a tor i n de p e n de n ce , a n d e qu i pm e nt c ons i st e n c y . I n de p e n de nt w e t - l ab v a l ida t i on da t a s e t ( ca ll ed wet_lab_validation_raw , n = 12) i n c lu di n g a sm a ll s e t o f r a n d omly ch os e n f ormul a t i ons r ec r ea t ed e xp e r i m e nt a lly to t e st t he pr edic t i v e p e r f orm a n ce o f m achi n e - l ea rn i n g mo de ls tr ai n ed on t he m ai n da t a s e t . R a w da t a s e t pr e pro ce ss i n g a n d c l ea n i n g A f t e r a pply i n g t he pr e pro ce ss i n g p i p e l i n e — i n c lu di n g checki n g p h ys ica l l i m i ts , r e mov i n g i nv a l id chi p c on di t i ons , c orr ec t i n g n a m i n g i n c ons i st e n cie s , a n d d ropp i n g r ed un da nt c olumns — t he c l ea n ed da t a s e t , ca ll ed formulations , c ont ai ns 304 da t a po i nts c orr e spon di n g to high - qu a l i ty f ormul a t i ons o b t ai n ed f rom t he c om bi n a t i on o f t he pr e pro ce ss ed s eed da t a s e t a n d t he pr e pro ce ss ed c ross - l ab or a tory e xt e ns i on da t a s e t . T hi s da t a s e t ca n be us ed f or st a t i st ica l or r eg r e ss i on mo de ll i n g , m achi n e - l ea rn i n g t a s k s , a n a lys i s o f f ormul a t i on – p e r f orm a n ce tr e n d s , de s ig n - sp ace e xplor a t i on , a n d be n ch m a r ki n g i nv e rs e - de s ig n a ppro ache s . T he pr e pro ce ss i n g wor kf low i n c lu de s : . C olumn st a n da r di z a t i on a n d f uzzy - m a t chi n g c orr ec t i on 59 i n c ons i st e nt l abe ls a utom a t ica lly c orr ec t ed . C onstr ai nt check s us i n g t he m e t ada t a fi l e s , f or e x a mpl e v e r if y i n g PDI ∈ [0,1] a n d SIZE ≤ 10,000 5 outl ie rs de t ec t ed ; c orr e spon di n g e ntr ie s r e mov ed . C hi p agg r ega t i on A ll D ropl e t - chi p e ntr ie s ( n = 25) r e mov ed d u e to l ack o f l i posom e f orm a t i on . D ropp i n g r ed un da nt c olumns ID , ML , CHIP , OUTPUT . O pt i on a l a u g m e nt a t i on SMO T E (25 % ) G a uss ia n no i s e (25 % ) A ll pro ce ss i n g s c r i pts a r e a v ai l ab l e i n code/preprocessing_scripts.py . U s age
README.md 2025-12-15 5 / 5 L o adi n g t he da t a s e t i n P yt h on import pandas as pd df = pd.read_csv("data/formulations.csv") E xplor i n g t he da t a s e t i n P yt h on # Once the dataset has been loaded # Inspect the first rows df.head() # Summary statistics df.describe() # Check distributions of key variables df[['SIZE', 'PDI', 'TFR', 'FRR']].hist(figsize=(10, 6)) E x a mpl e o f a m achi n e - l ea rn i n g t a s k i n P yt h on from sklearn.model_selection import train_test_split from sklearn.ensemble import RandomForestRegressor # Once the dataset has been loaded # Divide targets from features X = df[['ESM','HSPC','CHOL','PEG','TFR','FRR']] y = df['SIZE'] # Perform train-test split X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2) # Try a simple regression model model = RandomForestRegressor() model.fit(X_train, y_train) # Check the results print("R squared:", model.score(X_test, y_test))
