cBioPortal Symposium Slides

Challenges and opportunities in the European cancer research data landscape Gerrit Meijer, MD, PhD, CSO Health-RI, pathologist & senior group leader NKI Lifang Liu, MD, PhD, Oncology node coordinator, Health-RI, Guest researcher NKI The mission of pathology 3Pathology: Hub and Integrator of Modern, Multidisciplinary [Precision] Oncology, Clin Cancer Res 2022 To obtain as much information as possible, from tissue-, cell-, and DNA-samples that is relevant for the best treatment of a patient The mission of pathology 4Pathology: Hub and Integrator of Modern, Multidisciplinary [Precision] Oncology, Clin Cancer Res 2022 To obtain as much information as possible, from tissue-, cell-, and DNA-samples that is relevant for the best treatment of a patient need for data! Holy trinity of data items in health research PatientCell line A learning healthcare system for getting to better health outcomes for citizens and patients faster DATA WORKING TOGETHER research evidence guideline change in practice Better health Health challenges 6 More & beter data to facilitate researchers to faster do more & better practice changing studies! Oneliners •Infrastructure is not a goal in itself, but a means to an end •The most useful infrastructure is the one that is actually being used •The better is the enemy of the good •Researchers and researchinfrastucture developers often seem to be living in parallel universes 7 Personalized medicine &health Personalized medicine: drugs CROs etc. Mission Better health for citizens and patients by reusing health & life sciences data with an integrated data infrastructure to enable data driven research, innovation and policy Health-RI: the Dutch national health datainfrastructure for research, innovation and policy making € 69M 2021-2028 Founded by multiple predecessors We basically only have three types of questions to healthcare data… Are we doing the right things? Healthcare evaluation and appropriate use Are we doing things the right way? Quality assurance Are new things better than the existing ones? Research & innovation … but hundreds of systems! 17 The problem Data cannot be linked at a personal level Data sets cannot be combined Data is not reusable Data is hard to find Data is inaccessible The challenge? Fragmentation! Not so much a technical problem, but rather an organizational, social, and cultural problem Using scientific research for solving societal challenges requires a paradigm shift Urgent need for new approaches: from craftsmanship to industrial scale 20 Health-data reuse obstacle-removal-trajectory Common interest General principes / Rating support / Communication Patients/citizens are not well informed about the importance of re-use of health(care) data Lack of clarity about who has control over the health(care) data Fear of lack of recognition by organisations generating health(care) data Access to health(care) data by/from companies can be particularly challenging Rules Technology put in practice Framework: Legal / Ethical / Societal / Privacy / Knowledge Security (Perceived) Legal / ethical / societal barriers of re-use of health(care) data Health(care) data cannot be linked to other data securely and precisely Safeguarding privacy Safeguarding knowledge security Multiple and varying review procedures for reuse of health(care) data Difficult to share health(care) data internationally Technical / Logistics / Organisation / Services Health(care) data are difficult to obtain and fragmented Health(care) data are changed and/or deleted (due to administrative burden) Health(care) data are not structured according to standards Technical / logistic / organisational (compliance-by-design) barriers Continous interaction Action program of ministries of health, science, economic affairs & Health-RI "What do we want with secondary use?” ” Are we allowed to do so?” ”Are we able to do so?” National registries & data initiatives Whole system in the room! Health & LS Data Collective voice & advocacy Shared data services ELSI Federated architecture FAIR data collections & biobanks National health data catalog Disease domain communities Industry Citizens & Patients Regional health/ hospital networks EU initiatives Government Funders Research &Healthcare professionals Connecting health care & research International alignment Sustainable funding models Human capital Scaling up from NL to EU …. 23 …. same problems, but solutions at the horizon 24 Directorate General Communications Networks, Content and Technology (CONNECT) Directorate General Health and Food Safety (SANTE) Directorate General Research & Innovation (RTD) Home to national nodes of European research infrastructures Health-RI bundles national nodes of •BBMRI: Biobanks & Cohorts •ELIXIR: Life science data infrastructure •EATRIS: Translational Medicine etc. Courtesy EUHPP webinar on European Health Data Space (2/3): Secondary Use of health data (27 February, 2025) ... and aligns perfectly well with the Health-RI deliverables 36 Blueprint of Health Data Access Body “Digital Business Capabilities” ties it all together ... “Webshop”= catalogue + order/request Cross border gateway to Healthdata@eu Transparancy portal Opt out registry Secure Processing Envaironment = Data access & analysis environment Data sets Data ”preparation” environment Evaluation environment Europe data holders: hospitals cohorts studies etc data stewards citizens evaluation committees data holders data users /researchers data users researchers Citizen/patient executes consent choice Data holder publishes data sets in catalogue Data user selects data set in catalogue, files application for permit and pays fee HDAB and/or other authority evaluates application, issues and publishes permit Data holder loads data into Data Preparation Environment, after preparation data is loaded into Secure Processing Environment Data user analyzes data, prepares tables & graphs within SPE and publishes results Citizen/patient sees in Transparancy Portal data applications evaluated and results published Health & LS Data Collective voice & advocacy Shared data services ELSI Federated architecture FAIR data collections & biobanks National health data catalog Disease domain communities National registries & data initiatives Industry & Technology Citizens & Patients Regional health/ hospital networks EU initiatives Government Funders Research &Healthcare professionals Connecting health care & research International alignment Sustainable funding models Human capital Etc. 52 Domain nodes = capitalizing on existing data infrastructure activities within disease domains This includes: o Connecting with networks of experts, users, data holders, funders, and developers. o Scaling up existing best practices o Contribute to domain-specific (meta)data models. o Gaining insights into users needs. o Bridging the gap between researchers and infrastructure the a A r chi t ec h tu r e Impleme n t a on D a t a Se r ices n c olo y ode n c ology node W 1 n c ology node W 2 n c olo y ode n c ology node W 3 n c ology node W the a E SI H D AB I n t ern a onal D S the a o ain ode o ain ode the a the a o ain ode o ain ode the a the a o ain ode o ain ode the a 53 Health-RI Domain nodes securing scalable solutions 54 The Cancer Mission supports the creation of UNCAN.eu, a federated European cancer research data infrastructure ..... establishing National Cancer Data Nodes, by the scaling-up or improvement of existing national health data infrastructures and by fostering their links to the European Health Data Space infrastructures for primary and secondary data uses ONCOLOGY https://candle-project.eu/ CANDLE composition 56 Consortium Composition N Partners Data Infrastructure 15 Cancer Research 26 Cancer Care 6 Patients 1 Digital Health 8 Public Health 7 Involvement in initiatives EOSC4CANCER 13 UNCAN.CSA 4 canSERV 5 EHDS -related Projects 5 Member states 20 https://candle-project.eu/ HADEA & Three DGs are involved 57 ‘With the aim to support the implementation of the CANDLE project maximizing results and impacts, and providing direct up-to-date information of linked critical policies to the consortium, these additional comments are provided in a harmonised summary to the coordinator/consortium to address them accordingly. These comments have resultednfrom exchanges among the pertinent EC services, namely, between HaDEA (the granting authority) and three policy DGs (RTD, SANTE, CNECT) leading the linked policies. Policy Officers -contact persons for the policy DGs: - DG RTD - Angelo Solmini (Unit D1 Combatting Diseases) - DG SANTE - Irini Kessissoglou (Unit C1 Digital Health) - DG CNECT – Alexandra Wesolowska (Unit H3 Research Coordination)’ the a A r chi t ec h tu r e Impleme n t a on D a t a Se r ices n c olo y ode n c ology node W 1 n c ology node W 2 n c olo y ode n c ology node W 3 n c ology node W the a E SI H D AB I n t ern a onal D S the a o ain ode o ain ode the a the a o ain ode o ain ode the a the a o ain ode o ain ode the a 58 EU National Cancer Data Nodes DK SE etc. NL n=20 CANDLE in a nutshell 59 Data Users Patients / civil society Health care professionals Researcher s Research infrastructur e builders Public authoritie s UNCAN.EU platform ECPDC information portal Stakeholder engagement & policy dialogue EHDS compliance at national and EU level Maturation Models NCDN landscape analysis and exemplar nodes Inequalities, digital skill divide and training syllabi REDUCE THE BURDEN OF CANCER IN EUROPE Communication & Dissemination National Cancer Data Nodes Exemplar national implementation roadmaps CANDLE Resource Kit Network of NCDNs Users’ needs and solutions Guidelines to connect and technical alignment with UNCAN.eu and ECPDC Communication & Dissemination Topical Users (e.g. AI) CANDLE spin-out collaboration with major EU projects 60 https://cbioportal.org @cbioportal 2006 2008 2024 2025 January 2009: Mutual exclusivity calculations https://cbioportal.org @cbioportal 2006 2008 2024 2025 January 2009: “SVG Fingerprints” https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 November 2010: Start of regular news updates https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 November 2010: Start of regular news updates https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 2011: JJ Gao joins the team / First SU2C portal JJ Gao https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 2011: 4 studies, 600 weekly visits https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 2012: Cerami et al. manuscript Cited by >17,000 https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 2012: Remond meets JJ https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 2012/2013: Patient view and study view Patient View Study View https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 Patient View 2013: 26 studies, 3,000 weekly visits https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 Patient View 2019: Group Comparison and code refactoring Group Comparison https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 Patient View 2020: Webinars, 295 studies in cBioPortal https://cbioportal.org @cbioportal Weekly user visits 2006 2008 10,000 15,000 5,000 Patient View 2020-2023: Adoption by additional consortia https://cbioportal.org @cbioportal Weekly user visits 2006 2008 2024 2025 10,000 15,000 5,000 Source : GA3 Source : GA4 2023: de Bruijn et al. 2023 (GENIE BPC) https://cbioportal.org @cbioportal Weekly user visits 2006 2008 2024 2025 10,000 15,000 5,000 Source : GA3 Source : GA4 2025: Clickhouse transition Aaron Lisman https://cbioportal.org @cbioportal Weekly user visits 2006 2008 2024 2025 10,000 15,000 5,000 Source : GA3 Source : GA4 cBioPortal usage over time https://cbioportal.org @cbioportal Weekly user visits 2006 2008 2024 2025 10,000 15,000 5,000 Source : GA3 Source : GA4 cBioPortal usage over time Technical glitches https://cbioportal.org @cbioportal Weekly user visits 2006 2008 2024 2025 10,000 15,000 5,000 Source : GA3 Source : GA4 cBioPortal usage over time Holidays https://cbioportal.org @cbioportal Weekly user visits 2006 2008 2024 2025 10,000 15,000 5,000 Source : GA3 Source : GA4 cBioPortal usage over time Curated Studies 493 Monthly Users ~39K Total Citations >31k > 8,000,000 visits to cBioPortal.org since 2011 >98 institutional instances globally Different cBioPortal instances Real World Data Sources Research Clinical Pathology Lab values Molecular Radiology Structured and NLP-derived data, incl. treatment & outcome Cell types, cancer types, genomic alterations Data Infrastructure Blood markers MSK-IMPACT clinical sequencing data Automated lesion detection, volumetric analysis, coregistration Data Abstraction & Interpretation Data Analytics Tools Outcome Models MSK Community Patients Raw Operational Data Research-Ready Data Scientific & Clinical Insights Automated Abstraction and Integration of Multimodal Real-World Data The Cancer Data Science Initiative at MSK Automated daily updates Main bottleneck: Manual clinical data abstraction 1. Train/validate NLP models on BPC-MSK cohort 2. Deploy NLP models on nonBPC-MSK cohort Clinical Data Mining Leverage curated data set to train & validate NLP methods Justin Jee Chris Fong Michele Waters Mirella Altoe Karl Pichotta Thinh Tran Tom Fu MSK-IMPACT clinical sequencing cohort GENIE BPC MSK-CHORD: Clinicogenomic Harmonized Oncologic Real-World Dataset ●NLP pipelines for clinical data abstraction ●Genomic data from MSK-IMPACT ●Enabling outcome models Jee et al. Nature (2024) © 2025 Memorial Sloan Kettering Cancer Center, et al. All rights reserved. MSK-CHORD at MSK: cBioPortal cohort with daily updates © 2025 Memorial Sloan Kettering Cancer Center, et al. All rights reserved. MSK-CHORD at MSK: cBioPortal cohort with daily updates ●Improved Clickhouse integration ●Imaging integration ○Pathology ○Radiology ○Image-derived data / embeddings ●AI query functionality ● … What is next for cBioPortal? cBioPortal Contributors Ugur Dogrusoz Ziya Erkoc Nikolaus Schultz Benjamin Gross S Onur Sumer Hongxin Zhang Ritika Kundra Ino de Bruijn Robert Sheridan Angelica Ochoa Aaron Lisman Manda Wilson Avery Wang Calvin Lu Anusha Satravada Ramyasree Madupuri Gaofei Zhao Xiang Li Anusha Satraveda Bryan Lai Rima AlHamad Jason Hwee Ethan Cerami Chris Sander Tali Mazor James Lindsay Jeremy Easton-Marks Ryan Fu Trevor Pugh Prasanna K Jagannathan Fedde Schaeffer Oleguer Plantalech Pim van Nierop Sander Rodenburg Mirella Kalafati Jessica Singh Matthijs Pon Tim Kuijpers Adam Resnick Allison Heath John Maris Charles Haynes B. Arman Aksoy Istemi Bahceci Caitlin Byrne Hsiao-Wei Chen Ersin Ciftci Fred Criscuolo Leonard Dervishi Gideon Dresdner Andy Dufilie Catherine Del Vecchio Fitz Arthur Goldberg Sjoerd van Hagen Zachary Heins Michael Heuer Anders Jacobsen Ngoc Nguyen Karthik Kalletla Yichao Sun Alexandros Sigaras Erik Larsson Dong Li Tamba Monrose Peter Kok Irina Pulyakhina Pichai Raman M. Furkan Sahin Kaan Sancak Sander de Ridder Sander Tan Paul van Dijk Jiaojiao Wang Stuart Watt James Xu Dionne Zaal Riza Nugraha Adam Abeshouse Luke Sikina Alumni Funding: Present & past Jianjiong Gao Priti Kumari Corey Dubin Pieter Lukasse Ruslan Forostianov Chris Sander Augustin Luna cBioPortal Contributors AcademicCommercial Pharma & biotech clients User experience: Centralized multiomics exploration using cBioPortal Sam de Vos (bioinformatician) cBioPortal Use cases in the Princess Maxima Center •Exploration of available biobank data •Preliminary analyses; hypothesis testing, extrapolation of findings. •Full data accessible through data request cBioPortal Considerations for choosing the platform •User-friendly; no steep learning curve •Active community •Fits the purpose: preliminary data exploration for all types of users cBioPortal Data release strategy •Monthly full biobank releases; static ”snapshots” of data available for requests •Data derived from automated ETL workflow cBioPortal Data release strategy •Monthly full biobank releases; static ”snapshots” of data available for requests •Data derived from automated ETL workflow •Release notes describing changes compared to last release cBioPortal Data content •Bulk RNA-Seq (Fusion, expression) •Somatic WGS (sSNVs, sCNVs, sSVs) •High level clinical data (ICCC/ICD-O) cBioPortal Data content •Bulk RNA-Seq (Fusion, expression) •Somatic WGS (sSNVs, sCNVs, sSVs) •High level clinical data (ICCC/ICD-O) cBioPortal Current developments: Structural variants •Visualization through Chromoscope implementation •Integrated view with CNVs/SNVs •New: integration in tabular view! Credits: Zhaoyuan Fu (Ryan), Dana Farber Cancer Institute cBioPortal Current development: Multi-institutional instance cBioPortal Hurdles and weaknesses •Content-wise: gene centricity excludes non-coding regions •Schema-dependent ingestion; merely a visualization endpoint •Striking a balance between completeness and intuitiveness; what is “preliminary analysis”? Acknowledgements Big Data Core Alex Janse Alex Staritsky Hinri Kerstens Patrick Kemmeren Diagnostic lab Clinicians & research nurses Trial & Data Center Biobank Project Team Funding IDT research IDT Data Intelligence team The Hyve Oleguer Plantalech Floris Vleugels Jessica Singh Marinel Cavelaar Parents & children Disqover team Menno de Vries Ingrid Schut Pieter Broers Bart Koppers Titel (2 regels) + subtitel LIJSTNIVEAU KIEZEN 1•Bullet 2 Gebruik onder de tab ‘Start’ de lijstniveau-knoppen om een tekst niveau te kiezen. Kies uit: •Sub-bullet #1 3–Sub-bullet #2 4Leestekst 5Kop #1 6 1. Numerieke bullet 7 a. Alfabetische bullet 8 Bronvermelding 9 Kop #2 MEER WETEN? Ga naar de instructie dia ‘VIDEO INSTRUCTIES’. Deze vind je vooraan de presentatie of voeg je in via ‘Start’ > ‘Nieuwe dia’ Start Lijstniveau verlagen Lijstniveau verhogen Collection and use of medical data Prospectief Landelijk CRC Cohort (PLCRC) Consent Unique Design Prospective Dutch Colorectal Cancer Cohort Study (PLCRC) All patients diagnosed with colorectal, small bowel, anal cancer at any time after diagnosis, all stages: Obversational studies: Clinical data Tissue Blood PROM’s Interventional studies: Trials within cohorts (TwiCs) Informed Consent Patient I would like to help future patients, for example my own children, that is why I participate. Participant PLCRC Use and storage of tissue Use and storage of blood tubes Participation in questionnaires (QoL) To be approached for new studies > 90% ~ 95% ~ 84% ~ 81% ~ 78% Alleen titel LIJSTNIVEAU KIEZEN 1•Bullet 2 Gebruik onder de tab ‘Start’ de lijstniveau-knoppen om een tekst niveau te kiezen. Kies uit: •Sub-bullet #1 3–Sub-bullet #2 4Leestekst 5Kop #1 6 1. Numerieke bullet 7 a. Alfabetische bullet 8 Bronvermelding 9 Kop #2 MEER WETEN? Ga naar de instructie dia ‘VIDEO INSTRUCTIES’. Deze vind je vooraan de presentatie of voeg je in via ‘Start’ > ‘Nieuwe dia’ Start Lijstniveau verlagen Lijstniveau verhogen 0 5000 10000 15000 20000 Inclusion PLCRC With the patient’s consent, we can use data and body material for research and policy questions •PLCRC open in ~60 hospitals •>30 hospitals collect blood for ctDNA studies •Blood has been collected from >5000 patients Afbeeldingsresultaat voor dccg ORCA CAIRO5 Samples Data? TCGA meeting 2012 Center for Translational Molecular Medicine Translational Research IT infrastructure: the TraIT road towards data integration and sustainability Remond J.A. Fijneman1, Sanne Abeln2, Guido Jenster3, Bauke Ylstra4, Connie R. Jimenez5, Jan-Willem Boiten6, Jeroen A.M. Belien1, Gerrit A. Meijer1 1Tumor Profiling Unit, 4Microarray Core Facility; 5Oncoproteomics Laboratory; VUmc Cancer Center Amsterdam, The Netherlands. [email protected] 2Centre for Integrative Bioinformatics VU (IBIVU), VU University Amsterdam; 3Josephine Nefkens Institute, Erasmus MC, Rotterdam; 6Center for Translational Molecular Medicine, High Tech Campus, Eindhoven, The Netherlands WP4 Molecular Profiling & User Stories: Analysis of molecular profiling data: Platform: Next Generation Sequencing (NGS) Leader: Guido Jenster User story: Identify and charaterise Fusion Genes in Prostate Cancer Platform: Arrays (DNA and RNA microarrays) Leader: Bauke Ylstra User story: Determine prognostic value of marker X in stage Y CRC Platform: Proteomics (mass spectrometry) Leader: Connie Jimenez User story: Identify and validate CRC protein biomarkers in stool Platform: Non-High-Throughput Molecular Profiling (NHTMP, ‘non-omics’) Leader: Remond Fijneman User story: Determine prognostic value of marker X in stage Y CRC Archival storage of molecular profiling data, allow to query: Platform: All of the above (platform-independent) Leader: Sanne Abeln User story: Data output formats, data storage needs, connectivity to core infrastructure (WP5) TraIT WP4 status and discussions •End-user driven use cases have been defined •Multidisciplinary team: biologists –platform specialists - (bio)informaticians •First iteration of tool assessments for data analysis by Q4 of 2012 Lessons to be learned: sets the example? •Data output formats •Data storage needs •Data access •Logging and version control of bioinformatics tools and pipelines •Integration of various molecular profiling data WP4 Objectives: What we aim to do •Improve Quality Assurance / Quality Control of research data acquisition •Metadata; logging samples, data capture, data storage, and data processing •Standardization of data storage (allow querying, allow interoperability) WP4 objectives logging samples QA/QC data capture data processing data storage allow to query Enable Research Introduction Projects in translational research share a common design, but also face very similar IT issues: •an explosion of data; •lack of integrated approaches for collection, management, and analysis of data; •lack of tools and infrastructure to collaborate with colleagues in a simple and secure manner. Workflow analysis and tool selection When further detailing the translational research workflow we still see a common workflow across the disease areas to be supported by TraIT: This workflow has been mapped onto the caBIG platform sponsored by the National Cancer Institute (NCI) as well as other promising tools like OpenClinica and TranSMART. All candidate tools are carefully evaluated in an assessment phase before adoption in TraIT. TraIT objectives It is TraIT’s ambition to provide a research IT platform that is: •Sustainable •Professionally supported •Endorsed by a strong community •Integrated across translational research domains •Supporting multi-center data sharing and analysis Division in work packages TraIT has been subdivided into four work packages (WPs) supporting data generating domains, and two work packages dealing with the overarching TraIT requirements: data integration and professional support respectively: Four data generating work packages Data integration & analysis across the four platforms Shared service center for hardware, training & support JJ Gao The PLCRC-DOLPHIN study DOLPHIN data: Patient information – blood samples – CT images Bl ood sampl e col l ect i on Imaging Pat i ent inclusion+ c li nica l overview CLIN. DATA Data management Data Collection cBioPortal Biobank Patient clinical data Data monitoring 1. Integration / Import script 2. cBioPortal formating script TGO - PLCRC SANDBOX PRODUCTION •Testing •Modeling •QC •Restrictive user access •Stable version •Updates •Users can request access Monthly* Monthly* CT scan images Seq. data MOL/IMG DATA DOLPHIN DATA WORKFLOW *not all data types 3. cBioPortal import script Images are being collected in XNAT-cBioPortal Bl ood sampl e col l ect i on Imaging Pat i ent inclusion+ c li nic a l overview Client CTP (Pseudonymisation) BMIA CTP Secure transfer connection Imaging data Inside Medical Centers DICOM protocol https://xnat.bmia.nl DICOM protocol https://cbioportal.health-ri.nl Resource linking eXtensible Neuroimaging Archive Toolkit Real time study monitoring with cBioPortal Blood collection Inclusion CT scan cBioPortal: Dynamic Data Export Challenge: Data slices cannot be easily taken out and shared Our solution: RFC95 - Study Export Ready! cBioPortal: Granular Access Permissions Challenge: Selective data sharing is difficult since access can only be controlled at the study level. Our solution: RFC96 - Authorization Model for Virtual Studies In progress… cBioPortal: Challenges and Future Improvements cBioPortal AI chat that navigates you through the site and answers analytical questions New analytical DB and importer for faster querying and ingestion. Summary Our work has taken us closer to expanding patient search, real time, across borders by improving: ●Connectivity: Sharing data (export/import) ●Data Management: Live data views and granular permissions Let's connect! We want to hear your challenges and exchange ideas on making data sharing seamless and secure. SE4BIO intro - We are part of steering committee - Deciding features and future - Brain and muscle Goal / story ● In cBioPortal, be able to analyze a patient’s data in the context of other patients, including patients that are in other hospitals across the nation ○ Very usual scenario when dealing with rare diseases ● Current limitations: ○ cBioPortal usually stand alone and available within institute ○ Aggregating data from collaborating institutions is limited to de-identified data, while local patients should be identifiable ○ Implementation The Future: LLM based Apps for patient discovery There are multiple different LLM applications for cBioPortal: - cBioPortal NavigationBot - AI-OQL - Model Context Protocol (Natural language to SQL, navigation,...) Consideration on WGS and cBioPortal •The interface & community we know and love •APIs / R / Python bindings •Open targets & OncoKB integration •Built for gene centric analysis (diffcult to do genomics, e.g. arm level events) •Data model has a variant type focus •No non coding / GWAS / RNA non-response biomarkers Combining genomic biomarkers to guide immunotherapy in non-small cell lung cancer Van de Haar, Mankor, Clinical Cancer Research Costs societal impact Clinical utility? Refinement Confirmation •20%, treatment could potentially be omitted* •Treatment cost savings ~€ >1.5 M •Screening all patients by WGS € <0.5 M *Prospective study in preparation at NKI/AVL Systematic searches for treatment + marker What that looks like ‘underneath’ Melanoma patients on immune therapy All melanoma patients by B2M alteration - Treatment filter - Less precise mutational filter Statistical power is limiting Cancer is common, but a specific type with a specific treatment is rare So we need to combine data https://hutch.health/relay GA4GH OMOP + https://eosc4cancer.eu/federatedcancer-data-analysis/ What we have for now Vignettes; example for NETs Pancreatic NET vs other NETs https://www.hartwigmedicalfoundation.nl/en/data/vignettes/ Germline Immune escape Genome wide features A comprehensive driver table gene category Likelihood Method Driver Likelihood missense nonsense splice frameshift inframe biallelic Min Copy Number Max Copy Number APC TSG DISRUPTION 0 0 0 0 0 0 0 1.38 2.97 FGFR2 ONCO DISRUPTION 0 0 0 0 0 0 0 2.01 4.01 GRIN2A TSG DISRUPTION 0 0 0 0 0 0 0 0.91 1.99 AMER1 TSG DNDS 0 1 0 0 0 0 0 2.02 2.02 CD79B ONCO AMP 1 0 0 0 0 0 0 6.13 6.13 PTEN TSG DEL 1 0 0 0 0 0 1 0.05 0.05 - 2015 - 2019 - 2022 - 2025 - 2026 From fat data to big data •expand databases by orders of magnitude •specific focus on rare cancers •AI approaches •multi-omics •multimodal (imaging, histopathology, etc) - 2024 International collaborations are key •connecting similar resources world-wide •global network based on coalition of the willing •uniformly analyzed genomics data •standardized clinical data •trusted authentication and federated access mechanisms - 2030 Ongoing projects - Genomics England (cross-validation) - DKFZ/NCT Germany (rare cancers) - UMCCR, Melbourne Australia - France Genomique (CUP) - Harmonisation of peadetric cancers Next steps? Example Request for Comments (RFCs) Documents ●RFC80 - Improve performance of cBioPortal Study View ●RFC90 - Extending Pairwise Comparison Plots Example Implementation RFC 90 Roadmap Roadmap LLM Integration Chat with data in cBioPortal’s ClickHouse database MCP MSK-CHORD Summary MCP MCP Model Context Protocol! Model Context Protocol Functions list_databases list_tables run_select_query Now, how to comprehensively test these results for accuracy?? Michele Waters et al. 100 Q/A pairs for AWS Imagine Grant Potential Integration in Website Imaging Integration AI-Driven Multimodal Integration Offers Many Exciting Opportunities Kevin Boehm, Sohrab Shah, Francisco Sanchez-Vega, manuscript in preparation. “Patient Maps” for Multimodal Data in cBioPortal “Patient Maps” for Multimodal Data in cBioPortal Similar approaches might work for Single Cell Data Adding Image Viewers and Derived Features Integrating spatial profiles and cancer genomics to identify immuneinfiltrated mismatch repair proficient colorectal cancers Technical Intro Treatment Naive (66/74) Colorectal Resections (74 patients) ~300 genes Acknowledgments NG-CHM MusicaTK