OncoGPT: A Modular AI Assistant Orchestrating LLMs in Molecular Oncology

OncoGPT: A Modular AI Assistant Orchestrating LLMs in Molecular Oncology | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article OncoGPT: A Modular AI Assistant Orchestrating LLMs in Molecular Oncology François Degrave, Cédric Balsat, Maxime Liénard, Sébastien Sauvage This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7869170/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract General-purpose large language models (LLMs) show promise for biomedical reasoning but remain ill-suited to regulated clinical workflows: they hallucinate, rely on opaque sources, and are difficult to audit—limitations incompatible with validated molecular reporting pipelines. A common response is to train or host domain-specific LLMs, yet this requires substantial data, infrastructure, and time. We present OncoGPT, a modular, provider-agnostic orchestration layer that enables the safe and auditable use of off-the-shelf LLMs in molecular oncology with minimal integration cost. A pluggable ModelSelector routes each query to on-premise or API models based on declarative capability and cost profiles, avoiding vendor lock-in and enabling model swaps by configuration rather than code. A hierarchical ContextBuilder assembles task-specific information so that outputs prioritize content from the injected context (e.g., report sections and linked references), with optional fallback to general biomedical knowledge when needed. Evaluated on 19 representative clinical prompts derived from real-world oncology reports, automatic model selection with context achieved expert acceptance across all prompts while reducing inference cost by an order of magnitude; by contrast, a fixed high-end model produced higher cost and lower expert-rated quality. These results demonstrate that a context-first, plug-and-play orchestration approach can operationalize general LLMs for traceable, cost-efficient support in precision oncology workflows—without training new domain-specific models. Biological sciences/Cancer Biological sciences/Computational biology and bioinformatics Health sciences/Health care Physical sciences/Mathematics and computing Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7869170","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":541684278,"identity":"e747e2d8-a4df-405d-98f3-b9c75306a5c7","order_by":0,"name":"François Degrave","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2UlEQVRIiWNgGAWjYBAC9gYGAzgLCJgJa+E5ANMCZJGqRSKBWC0MzBsfft1jl9gv+fjYgx8V1nn8DcwPH93Aq4Wt2FjmWXLizNlp6YY9Z9KLJQ6wGRvn4NFiz8BjJi1xgNnY4HaOmTRj2+HEhgM8bNL4tPAw8Jj/ljhQb2x/8wxEy3witJgxfjhwWM5AggeiZQNBLcxsxdIMB47LSZxJS5ME+iVx42ECfuFhb9748ceBah7+9sPHJIAhljjvePPDx/i0gCKCmQddhCBg/EGEolEwCkbBKBjBAAAeAUa2pkrarQAAAABJRU5ErkJggg==","orcid":"","institution":"OncoDNA SA","correspondingAuthor":true,"prefix":"","firstName":"François","middleName":"","lastName":"Degrave","suffix":""},{"id":541684279,"identity":"64be631c-c184-45c4-bcd8-9721e79bf7d7","order_by":1,"name":"Cédric Balsat","email":"","orcid":"","institution":"OncoDNA SA","correspondingAuthor":false,"prefix":"","firstName":"Cédric","middleName":"","lastName":"Balsat","suffix":""},{"id":541684280,"identity":"597e5639-53bd-473a-a660-ad7028b5cfcd","order_by":2,"name":"Maxime Liénard","email":"","orcid":"","institution":"OncoDNA SA","correspondingAuthor":false,"prefix":"","firstName":"Maxime","middleName":"","lastName":"Liénard","suffix":""},{"id":541684281,"identity":"304d4b16-d7d3-49af-a0ce-721aa913bf43","order_by":3,"name":"Sébastien Sauvage","email":"","orcid":"","institution":"OncoDNA SA","correspondingAuthor":false,"prefix":"","firstName":"Sébastien","middleName":"","lastName":"Sauvage","suffix":""}],"badges":[],"createdAt":"2025-10-15 14:26:01","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7869170/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7869170/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":95715369,"identity":"2339dd34-015b-4f6c-aa9c-a83d233dcf9a","added_by":"auto","created_at":"2025-11-12 08:39:39","extension":"json","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6931,"visible":true,"origin":"","legend":"","description":"","filename":"de4e329878854a60bf7b431ced4b2f3e.json","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/028f0299c4408e838c751d54.json"},{"id":95715371,"identity":"b892dbc0-9612-4eb0-808f-7a875235a27a","added_by":"auto","created_at":"2025-11-12 08:39:40","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16204,"visible":true,"origin":"","legend":"","description":"","filename":"answerqualitybarplot.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/15c4129091398adb69df01c0.pdf"},{"id":95715365,"identity":"29b6dbd6-30fd-4eab-84ad-4ec53c85a512","added_by":"auto","created_at":"2025-11-12 08:39:38","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":65560,"visible":true,"origin":"","legend":"","description":"","filename":"coverletter.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/e95a30ef1d6c763bf7bc3972.pdf"},{"id":95715332,"identity":"bf6be408-74f7-497c-a804-f6c9c3919040","added_by":"auto","created_at":"2025-11-12 08:39:35","extension":"pdf","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16306,"visible":true,"origin":"","legend":"","description":"","filename":"modelusageoncogptpie.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/eec184d2b9fe0e88fa5e0774.pdf"},{"id":95715331,"identity":"026a561b-a87a-481d-a21b-8ac2cd3e7fb9","added_by":"auto","created_at":"2025-11-12 08:39:34","extension":"pdf","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":286443,"visible":true,"origin":"","legend":"","description":"","filename":"oncogptarticle.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/151f3ba9a9968499c18d7d88.pdf"},{"id":95715370,"identity":"d3f46c4b-feb7-435d-9c76-fa09271504d6","added_by":"auto","created_at":"2025-11-12 08:39:39","extension":"pdf","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15884,"visible":true,"origin":"","legend":"","description":"","filename":"sbertstabilitydistribution.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/ed54826fb3141a5e49699b93.pdf"},{"id":95715366,"identity":"e627c927-ba71-450d-8702-abaee6eeaf6d","added_by":"auto","created_at":"2025-11-12 08:39:38","extension":"pdf","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15168,"visible":true,"origin":"","legend":"","description":"","filename":"stabilitydistribution.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/902cb34b46c7589c7ab9cd7d.pdf"},{"id":95715367,"identity":"f399c388-b7c3-48e7-8410-a58933c77bd4","added_by":"auto","created_at":"2025-11-12 08:39:38","extension":"pdf","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18237,"visible":true,"origin":"","legend":"","description":"","filename":"strategyvsintentheatmap.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/1b7869689fdd17f2e8926134.pdf"},{"id":95715368,"identity":"9339c81d-b3d5-44bc-9a41-470a7caf177d","added_by":"auto","created_at":"2025-11-12 08:39:38","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":87311,"visible":true,"origin":"","legend":"","description":"","filename":"de4e329878854a60bf7b431ced4b2f3e1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1/8fc76a1f4b06f56d3b08756b.xml"},{"id":102313579,"identity":"0764a4dd-51c1-4981-9cd8-c11e689bac04","added_by":"auto","created_at":"2026-02-10 12:12:48","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":273874,"visible":true,"origin":"","legend":"","description":"","filename":"oncogptarticle.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7869170/v1_covered_819396fe-845a-44aa-bc68-4f113a4adac2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"OncoGPT: A Modular AI Assistant Orchestrating LLMs in Molecular Oncology","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7869170/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7869170/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGeneral-purpose large language models (LLMs) show promise for biomedical reasoning but remain ill-suited to regulated clinical workflows: they hallucinate, rely on opaque sources, and are difficult to audit—limitations incompatible with validated molecular reporting pipelines. A common response is to train or host domain-specific LLMs, yet this requires substantial data, infrastructure, and time.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe present OncoGPT, a modular, provider-agnostic orchestration layer that enables the safe and auditable use of off-the-shelf LLMs in molecular oncology with minimal integration cost. A pluggable\u0026nbsp;ModelSelector routes each query to on-premise or API models based on declarative capability and cost profiles, avoiding vendor lock-in and enabling model swaps by configuration rather than code. A hierarchical\u0026nbsp;ContextBuilder assembles task-specific information so that outputs prioritize content from the injected context (e.g., report sections and linked references), with optional fallback to general biomedical knowledge when needed. Evaluated on 19 representative clinical prompts derived from real-world oncology reports, automatic model selection with context achieved expert acceptance across all prompts while reducing inference cost by an order of magnitude; by contrast, a fixed high-end model produced higher cost and lower expert-rated quality. These results demonstrate that a context-first, plug-and-play orchestration approach can operationalize general LLMs for traceable, cost-efficient support in precision oncology workflows—without training new domain-specific models.\u003c/p\u003e","manuscriptTitle":"OncoGPT: A Modular AI Assistant Orchestrating LLMs in Molecular Oncology","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-12 08:38:31","doi":"10.21203/rs.3.rs-7869170/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"66a6c2c0-ffa1-41c6-952b-c25af126049d","owner":[],"postedDate":"November 12th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":57633154,"name":"Biological sciences/Cancer"},{"id":57633155,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":57633156,"name":"Health sciences/Health care"},{"id":57633157,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2026-02-10T12:12:25+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-12 08:38:31","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7869170","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7869170","identity":"rs-7869170","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}