Cluster-guided multimodal deep learning improves gastric cancer prognosis prediction and uncovers interpretable immune phenotypes

Cluster-guided multimodal deep learning improves gastric cancer prognosis prediction and uncovers interpretable immune phenotypes | 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 Cluster-guided multimodal deep learning improves gastric cancer prognosis prediction and uncovers interpretable immune phenotypes Chen Su, Jingjing Li, Yuhang Liu, Changhang Lin, Jingjing Hou, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9575924/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 8 You are reading this latest preprint version Abstract Recent advances in multimodal deep learning have significantly improved tumor survival analysis. However, gastric cancer exhibits high molecular heterogeneity, and conventional universal models fail to capture subtype-specific prognostic signals. Here, we show that mACE, a multimodal attention clustering expert framework, stratifies patients into biologically homogeneous immune subtypes via deep embedding clustering of fused histopathological and transcriptomic representations, then trains independent expert models per subtype. Evaluated across four independent cohorts, mACE achieved a mean C-index of 0.769, outperforming stratified and random sampling baselines by an average of 7.7%. Spatial attention analysis confirmed concordance between model focus and per-cluster pathway biology, providing direct histopathological interpretability. Enrichment analysis further revealed clinically actionable dissociations invisible to existing molecular classifiers: a subset of MSI-H tumors harboring immune-cold microenvironments despite high mutational burden, and genomically stable tumors exhibiting immune-hot phenotypes representing a hidden responder population overlooked by TMB-based selection. This work establishes biologically informed AI stratification as a scalable paradigm for precision tumor prognosis and immunotherapy patient selection. Health sciences/Biomarkers Biological sciences/Cancer Biological sciences/Computational biology and bioinformatics Biological sciences/Immunology Health sciences/Oncology Gastric cancer Computational pathology Whole slide image Survival analysis Immune subtype Multimodal deep learning Bioinformatics Full Text Additional Declarations No competing interests reported. Supplementary Files ExtendedDataFigures.pdf SupplementaryTable.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 17 May, 2026 Reviewers agreed at journal 08 May, 2026 Reviewers agreed at journal 06 May, 2026 Reviewers agreed at journal 06 May, 2026 Reviewers invited by journal 06 May, 2026 Editor assigned by journal 04 May, 2026 Submission checks completed at journal 04 May, 2026 First submitted to journal 30 Apr, 2026 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-9575924","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":638609333,"identity":"e7bd8017-ff72-4763-a518-98e0ab4edcf5","order_by":0,"name":"Chen Su","email":"","orcid":"","institution":"Zhongshan Hospital of Xiamen University","correspondingAuthor":false,"prefix":"","firstName":"Chen","middleName":"","lastName":"Su","suffix":""},{"id":638609335,"identity":"1daec189-bdff-454d-95a1-57a8093f41b8","order_by":1,"name":"Jingjing Li","email":"","orcid":"","institution":"Macao Polytechnic University","correspondingAuthor":false,"prefix":"","firstName":"Jingjing","middleName":"","lastName":"Li","suffix":""},{"id":638609337,"identity":"aecfe07b-d98a-4908-990d-7cf6b82cfa7d","order_by":2,"name":"Yuhang Liu","email":"","orcid":"","institution":"Macao Polytechnic University","correspondingAuthor":false,"prefix":"","firstName":"Yuhang","middleName":"","lastName":"Liu","suffix":""},{"id":638609343,"identity":"c75f1bca-1042-4fe4-a10b-bca339de34b1","order_by":3,"name":"Changhang Lin","email":"","orcid":"","institution":"Macao Polytechnic University","correspondingAuthor":false,"prefix":"","firstName":"Changhang","middleName":"","lastName":"Lin","suffix":""},{"id":638609350,"identity":"2e1e3943-3fa8-440a-a7a0-4452a031ba4e","order_by":4,"name":"Jingjing Hou","email":"","orcid":"","institution":"Zhongshan Hospital of Xiamen University","correspondingAuthor":false,"prefix":"","firstName":"Jingjing","middleName":"","lastName":"Hou","suffix":""},{"id":638609352,"identity":"dbb77939-58dd-4e8c-b36c-9a8284fb9ce7","order_by":5,"name":"Shuwen Xiong","email":"","orcid":"","institution":"Macao Polytechnic University","correspondingAuthor":false,"prefix":"","firstName":"Shuwen","middleName":"","lastName":"Xiong","suffix":""},{"id":638609353,"identity":"e9c520d6-3daa-458f-ab20-2f567642768f","order_by":6,"name":"Linjie Qu","email":"","orcid":"","institution":"Xiamen University","correspondingAuthor":false,"prefix":"","firstName":"Linjie","middleName":"","lastName":"Qu","suffix":""},{"id":638609357,"identity":"ead72470-2086-42d4-8342-b4bfb112924d","order_by":7,"name":"Zeyang Lin","email":"","orcid":"","institution":"Zhongshan Hospital of Xiamen University","correspondingAuthor":false,"prefix":"","firstName":"Zeyang","middleName":"","lastName":"Lin","suffix":""},{"id":638609360,"identity":"e8296886-85bd-4f83-8a9e-4d88c72d3cae","order_by":8,"name":"Quan Zou","email":"","orcid":"","institution":"University of Electronic Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Quan","middleName":"","lastName":"Zou","suffix":""},{"id":638609362,"identity":"bb76c6a3-bcc1-45b0-8a91-011731c55b9a","order_by":9,"name":"Ran Su","email":"","orcid":"","institution":"Tianjin University","correspondingAuthor":false,"prefix":"","firstName":"Ran","middleName":"","lastName":"Su","suffix":""},{"id":638609363,"identity":"351994f6-02a3-4233-bd7f-f81caa73b308","order_by":10,"name":"Leyi Wei","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0UlEQVRIiWNgGAWjYFACxsYDDAwSchAOG3FaGkBajEnRwsAA1MKQ2EC0FoPjzQ0HPrZZpK+d3WPA8KHsMINuewMBLWcONhyc2SaRu+3OGQPGGecOM5idOYBfi9mNxIbDvCAtN3IMmHnbgFpuJBDQcv9hw+G/bRLpZiAtf4nScoOx4TBjm0QCWAsjMVrszyQ2HOw5J2G47c6xAiAjnYegXyTbjz988KOsTt7sdvNGIMNazux4A34tYMAIig4JSATxEKEeBP5AtIyCUTAKRsEowAoA+KBOFTs+NO0AAAAASUVORK5CYII=","orcid":"","institution":"Macao Polytechnic University","correspondingAuthor":true,"prefix":"","firstName":"Leyi","middleName":"","lastName":"Wei","suffix":""}],"badges":[],"createdAt":"2026-04-30 10:38:40","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9575924/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9575924/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109405261,"identity":"c8c397ba-adb8-49ea-91f4-67e6745ce758","added_by":"auto","created_at":"2026-05-17 13:14:54","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1304563,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9575924/v1_covered_6a716472-5f75-4e5c-8db7-adba6c32dff5.pdf"},{"id":109316358,"identity":"a4bb6842-9a8f-4b10-8718-e07499239d8f","added_by":"auto","created_at":"2026-05-15 12:26:49","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":849119,"visible":true,"origin":"","legend":"","description":"","filename":"ExtendedDataFigures.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9575924/v1/4328405069be57464d35f229.pdf"},{"id":109316359,"identity":"1701e850-23e5-4e21-ac56-d77fcab6655b","added_by":"auto","created_at":"2026-05-15 12:26:49","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":26987,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable.docx","url":"https://assets-eu.researchsquare.com/files/rs-9575924/v1/b9e4631f66dabb92134593cc.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Cluster-guided multimodal deep learning improves gastric cancer prognosis prediction and uncovers interpretable immune phenotypes","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Gastric cancer, Computational pathology, Whole slide image, Survival analysis, Immune subtype, Multimodal deep learning, Bioinformatics","lastPublishedDoi":"10.21203/rs.3.rs-9575924/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9575924/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eRecent advances in multimodal deep learning have significantly improved tumor survival analysis. However, gastric cancer exhibits high molecular heterogeneity, and conventional universal models fail to capture subtype-specific prognostic signals. Here, we show that mACE, a multimodal attention clustering expert framework, stratifies patients into biologically homogeneous immune subtypes via deep embedding clustering of fused histopathological and transcriptomic representations, then trains independent expert models per subtype. Evaluated across four independent cohorts, mACE achieved a mean C-index of 0.769, outperforming stratified and random sampling baselines by an average of 7.7%. Spatial attention analysis confirmed concordance between model focus and per-cluster pathway biology, providing direct histopathological interpretability. Enrichment analysis further revealed clinically actionable dissociations invisible to existing molecular classifiers: a subset of MSI-H tumors harboring immune-cold microenvironments despite high mutational burden, and genomically stable tumors exhibiting immune-hot phenotypes representing a hidden responder population overlooked by TMB-based selection. This work establishes biologically informed AI stratification as a scalable paradigm for precision tumor prognosis and immunotherapy patient selection.\u003c/p\u003e","manuscriptTitle":"Cluster-guided multimodal deep learning improves gastric cancer prognosis prediction and uncovers interpretable immune phenotypes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-15 12:26:45","doi":"10.21203/rs.3.rs-9575924/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-17T22:47:29+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"158973381426475179839576554500774676059","date":"2026-05-08T21:57:39+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"134336155804431465991915179404186022099","date":"2026-05-06T23:32:40+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"213385590504134472993051747057529871289","date":"2026-05-06T12:03:19+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-05-06T09:52:03+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-05T00:00:21+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-05-04T10:11:36+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Digital Medicine","date":"2026-04-30T10:33:30+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"cfa7ca5c-d3b7-4364-8f6e-b7cbb2bca005","owner":[],"postedDate":"May 15th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-17T22:47:29+00:00","index":20,"fulltext":""},{"type":"reviewerAgreed","content":"158973381426475179839576554500774676059","date":"2026-05-08T21:57:39+00:00","index":18,"fulltext":""},{"type":"reviewerAgreed","content":"134336155804431465991915179404186022099","date":"2026-05-06T23:32:40+00:00","index":15,"fulltext":""},{"type":"reviewerAgreed","content":"213385590504134472993051747057529871289","date":"2026-05-06T12:03:19+00:00","index":14,"fulltext":""},{"type":"reviewersInvited","content":"10","date":"2026-05-06T09:52:03+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-05T00:00:21+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-05-04T10:11:36+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Digital Medicine","date":"2026-04-30T10:33:30+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":67965801,"name":"Health sciences/Biomarkers"},{"id":67965802,"name":"Biological sciences/Cancer"},{"id":67965803,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":67965804,"name":"Biological sciences/Immunology"},{"id":67965805,"name":"Health sciences/Oncology"}],"tags":[],"updatedAt":"2026-05-15T12:26:46+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-15 12:26:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9575924","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9575924","identity":"rs-9575924","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}