{
  "abstract": "Immune checkpoint inhibitors are standard across cancers, yet most patients do not respond and existing biomarkers generalize poorly across tumor types, drugs and clinical settings. We present C ompass , a pan-cancer foundation model that predicts immunotherapy response from bulk tumor transcriptomes using a concept-bottleneck transformer. C ompass encodes gene expression through 44 biologically grounded immune concepts representing immune cell states, tumor-microenvironment interactions, and signaling pathways. Trained on 10,184 tumors across 33 cancer types, C ompass outperforms 22 baseline methods in 16 independent clinical cohorts spanning seven cancers and six immune checkpoint inhibitors, increasing accuracy by 8.5% and area under the precision-recall curve by 15.7%, with minimal additional training. The model generalizes to unseen cancer types and treatments, supporting indication selection and patient stratification in early-phase clinical trials. In survival analyses, C ompass -stratified responders have longer overall survival (hazard ratio = 4.7, p < 0.0001). Personalized response maps connect gene expression to immune concepts, revealing mechanisms of response and resistance; in immune-inflamed non-responders, C ompass highlights programs including TGF- β signaling, endothelial exclusion, CD4+ T cell dysfunction, and B cell deficiency. By combining interpretability with transfer learning, C ompass enables robust prediction and mechanistic insight to inform trial design and translational studies.",
  "abstract_shown": true,
  "affiliations": [
    "College of Pharmaceutical Sciences, Zhejiang University, Hangzhou, China",
    "Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA",
    "Division of Immunology, Boston Children's Hospital, Harvard Medical School, Boston, MA, USA",
    "Roche Pharma Research and Early Development, Oncology Early Clinical Development, Roche Innovation Center Basel, F. Hoffmann-La Roche Ltd., Basel, Switzerland",
    "Computational Sciences Center of Excellence, F. Hoffmann-La Roche Ltd., Basel, Switzerland. daniel.marbach.dm1@roche.com",
    "Broad Institute of MIT and Harvard, Cambridge, MA, USA. marinka@hms.harvard.edu",
    "Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA. marinka@hms.harvard.edu",
    "Harvard Data Science Initiative, Cambridge, MA, USA. marinka@hms.harvard.edu",
    "Kempner Institute for the Study of Natural and Artificial Intelligence, Harvard University, Allston, MA, USA. marinka@hms.harvard.edu"
  ],
  "article_type": null,
  "arxiv": null,
  "author_total": 8,
  "authors": [
    {
      "affs": [
        0,
        1
      ],
      "name": "Wan Xiang Shen",
      "orcid": "0000-0001-7114-3664"
    },
    {
      "affs": [
        1
      ],
      "name": "Intae Moon",
      "orcid": "0000-0002-0978-9605"
    },
    {
      "affs": [
        2
      ],
      "name": "Thinh H. Nguyen",
      "orcid": "0000-0001-6660-6556"
    },
    {
      "affs": [
        1
      ],
      "name": "Michelle M. Li",
      "orcid": "0000-0003-0223-7485"
    },
    {
      "affs": [
        1
      ],
      "name": "Yepeng Huang"
    },
    {
      "affs": [
        3
      ],
      "name": "Nitya Nair",
      "orcid": "0000-0002-9524-9313"
    },
    {
      "affs": [
        4
      ],
      "name": "Daniel Marbach",
      "orcid": "0000-0002-6686-9249"
    },
    {
      "affs": [
        5,
        6,
        7,
        8
      ],
      "name": "Marinka Žitnik",
      "orcid": "0000-0001-8530-7228"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/bc0zd0z703hgw.bib",
  "cited_by": 32,
  "doi": "10.1038/s41591-026-04502-7",
  "fulltext_url": null,
  "id": "bc0zd0z703hgw",
  "issns": [
    "1078-8956",
    "1546-170X",
    "1744-7933"
  ],
  "landing_url": "https://doi.org/10.1038/s41591-026-04502-7",
  "language": "en",
  "licence": "CC-BY-NC-ND",
  "licence_label": "CC BY-NC-ND",
  "licence_source": "journal",
  "licence_url": "https://creativecommons.org/licenses/by-nc-nd/4.0/",
  "metadata_licence": null,
  "native_subjects": [
    [
      "openalex",
      "2730|Cancer Immunotherapy and Biomarkers"
    ]
  ],
  "openalex": "W4410090477",
  "pdf_url": "https://www.nature.com/articles/s41591-026-04502-7.pdf",
  "pmcid": "PMC12083594",
  "pmid": "40385399",
  "published": "2026-07-03",
  "publisher": "Nature Portfolio",
  "ris": "https://research.draconitic.com/paper/bc0zd0z703hgw.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T13:26:54Z",
      "native_id": "W4410090477",
      "source": "openalex",
      "url": "https://doi.org/10.1038/s41591-026-04502-7"
    }
  ],
  "subjects": [
    {
      "field": "medicine",
      "origin": "openalex:2730|Cancer Immunotherapy and Biomarkers",
      "subfield": "oncology"
    }
  ],
  "title": "Generalizable AI predicts immunotherapy outcomes across cancers and treatments",
  "title_text": "Generalizable AI predicts immunotherapy outcomes across cancers and treatments",
  "type": "peer-reviewed",
  "updated_at": "2026-11-09T13:21:15Z",
  "url": "https://research.draconitic.com/paper/bc0zd0z703hgw",
  "venue": "Nature Medicine",
  "versions": []
}