Medea: an omics AI agent for therapeutic discovery that performs verified, long-horizon reasoning across protein contexts, cell states, and personalized patient omics data.<\/span><\/span><\/p>\n Medea validates each decision and output against data and tool constraints as the analysis unfolds. <\/a><\/p>\n Pengwei Sui<\/a> Michelle L.<\/a> Shanghua Gao<\/a><\/p>\n \ud83d\udc49 Medea: https:\/\/lnkd.in\/eGNy5Q_D<\/a><\/p>\n \ud83d\udc49 Open science: https:\/\/lnkd.in\/eNfp3D3B<\/a><\/p>\n \ud83d\udc49 Paper: https:\/\/lnkd.in\/efe4-98b<\/a><\/p>\n 2\ufe0f\u20e3 MEDEA is a verification-aware omics agent designed for long-horizon therapeutic reasoning. \u2022 Context + integrity verification while building the research plan (catches context slips early) 3\ufe0f\u20e3 We evaluate Medea across 5,679 analyses in three open-ended domains:<\/p>\n \ud83d\udd39 Target identification across cell type contexts 4\ufe0f\u20e3 Target discovery fails when biology is averaged. Medea reasons at cell-type resolution, integrating:<\/p>\n \u2022 single-cell atlases 5\ufe0f\u20e3\u00a0Synthetic lethality asks a subtle question: Do two gene perturbations together reduce viability more than expected from either alone?<\/p>\n Medea treats this as an evidence synthesis task, combining:<\/p>\n \u2022 CRISPR dependency data And it abstains when evidence is weak.<\/p>\n 6\ufe0f\u20e3 Across seven cancer cell lines, Medea outperforms LLMs in identifying true synthetic lethal interactions.<\/p>\n Medea not only predicts more positives but also corrects LLM errors and avoids confident false claims.<\/p>\n 7\ufe0f\u20e3 Immunotherapy response prediction from personal transcriptomes<\/p>\n Predicting immunotherapy response requires reasoning over:<\/p>\n \u2022 tumor transcriptomes 8\ufe0f\u20e3 Medea works where heuristics fail. Across TMB-high\/low and inflamed\/non-inflamed tumors, Medea:<\/p>\n \u2022 outperforms LLMs Congrats to @sui67713 @_michellemli @GaoShanghua Wanxiang Shen Valentina Giunchiglia Andrew Shen Yepeng Huang Zhenglun Kong<\/p>\n https:\/\/lnkd.in\/eGNy5Q_D<\/a><\/p>\n Open science: https:\/\/lnkd.in\/eNfp3D3B<\/a><\/p>\n https:\/\/lnkd.in\/efe4-98b<\/a><\/p>\n Harvard Medical School<\/a> Harvard University<\/a> Harvard Medical School Department of Biomedical Informatics<\/a> Broad Institute of MIT and Harvard<\/a> Kempner Institute at Harvard University<\/a> Harvard Data Science Initiative<\/a><\/p>\n<\/div>\n<\/div>\n<\/div>\n Marinka Zitnik \u2022 Following\u00a0Associate Professor at Harvard Medea: an omics AI agent for therapeutic discovery that performs verified, long-horizon reasoning across protein contexts, cell states, and personalized patient omics data. Medea validates each decision and output against data and tool constraints as the analysis unfolds. Pengwei Sui Michelle L. Shanghua Gao \ud83d\udc49 Medea: https:\/\/lnkd.in\/eGNy5Q_D \ud83d\udc49 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[24],"tags":[],"_links":{"self":[{"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/posts\/2268"}],"collection":[{"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/comments?post=2268"}],"version-history":[{"count":5,"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/posts\/2268\/revisions"}],"predecessor-version":[{"id":2276,"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/posts\/2268\/revisions\/2276"}],"wp:attachment":[{"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/media?parent=2268"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/categories?post=2268"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jonathandespres.com\/cryonicsrevival\/wp-json\/wp\/v2\/tags?post=2268"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\nWhat is unique:<\/p>\n
\n\u2022 Pre-run and post-run checks around tool execution (an analysis can run without errors and still be wrong)
\n\u2022 Study relevance + evidence-strength screening for literature (not just retrieval)
\n\u2022 Multi-round consensus that reconciles tool outputs, literature, and the LLM, with calibrated abstention when evidence is insufficient
\n\u2022 A rich tool space with single-cell and bulk transcriptomics FMs, genetic dependency maps, protein and tissue networks, pathway and ontology resources, and interpretable ML models<\/p>\n
\n\ud83d\udd39 Synthetic lethality reasoning in cancer cell lines
\n\ud83d\udd39 Immunotherapy response prediction from patient transcriptomes<\/p>\n
\n\u2022 disease genetics
\n\u2022 drug evidence<\/p>\n
\n\u2022 pathway context
\n\u2022 literature support<\/p>\n
\n\u2022 immune programs
\n\u2022 clinical context<\/p>\n
\n\u2022 rescues errors from ML models
\n\u2022 adapts reasoning to subgroup biology<\/p>\n<\/div>\n<\/div>\n
<\/div>\n<\/div>\n
<\/div>\n<\/div>\n
<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"