DrugSAGE: an aggregation-based method for drug response imputation

DrugSAGE: an aggregation-based method for drug response imputation | 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 Method Article DrugSAGE: an aggregation-based method for drug response imputation Peilin Jia, Zhongming Zhao This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8634723/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 13 You are reading this latest preprint version Abstract Accurate prediction of drug response is critical for optimizing the clinical application of cancer therapies. Nevertheless, owing to genetic and genomic heterogeneity and the complex tumor microenvironment, cancer samples may exhibit heterogeneous responses to the same therapeutic agents, even when harboring identical driver mutations or genes. In this study, we developed DrugSAGE, a novel framework that leverages Graph Neural Networks (GNNs) to model cellular heterogeneity and accurately predict drug response from transcriptomic data. DrugSAGE enhances prediction by aggregating features from a sample and its most similar counterparts. Critically, we introduce a customized linear layer incorporating gene-pathway annotations to provide biological interpretability, facilitating the identification of key pathways driving the prediction. We benchmarked DrugSAGE using TCGA bulk data, six public bulk datasets, and four scRNA-seq datasets. Predictions from DrugSAGE showed significant associations between drugs and their known target genes (e.g., HER2 + inhibitors, MET inhibitors, BRAF inhibitors) or significant differences between patient groups stratified by their treatments (e.g., Erlotinib, PLX4720, 5-Fluorouracil). Furthermore, we show DrugSAGE can effectively predict drug response at the single-cell level. Additionally, by analyzing pathway-based embeddings, we identified pathways crucial to the models, including those involved in cancer, MAPK signaling, endocytosis, and JAK-STAT signaling, among others. Compared to previous methods, DrugSAGE demonstrated superior or comparable performance, offering a novel approach for predicting drug responses. GraphSAGE Graph Neural Network drug response Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Full Text Additional Declarations No competing interests reported. Supplementary Files TableS1.xlsx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 24 Mar, 2026 Reviews received at journal 18 Mar, 2026 Reviews received at journal 13 Mar, 2026 Reviews received at journal 13 Mar, 2026 Reviewers agreed at journal 11 Mar, 2026 Reviewers agreed at journal 08 Mar, 2026 Reviews received at journal 07 Mar, 2026 Reviewers agreed at journal 06 Mar, 2026 Reviewers agreed at journal 06 Mar, 2026 Reviewers invited by journal 06 Mar, 2026 Editor assigned by journal 26 Jan, 2026 Submission checks completed at journal 19 Jan, 2026 First submitted to journal 18 Jan, 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-8634723","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Method Article","associatedPublications":[],"authors":[{"id":602732217,"identity":"981e310e-817f-415f-9ad8-2fd71389d03f","order_by":0,"name":"Peilin Jia","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAsElEQVRIiWNgGAWjYFCCgw0MHxskSNTCOJNELQwMzLwNpCg3Zzzc+Nl2h4VdA/vhBww/dxChxbLhYLN07hmJ5AaeNAPG3jNEaDE4cLBBOrdNIpmBIYeBmbGNOC3Nvy1BWvjfEK+lTZqxTcKOQYJYW4B+abPsPSORwCbxzOBgLzFazCWOP77xc0edPT9/8sMHP4lymMQBMJ0IUnyACA1ALfwNYNqeKNWjYBSMglEwMgEAQ7U2ZdyP78gAAAAASUVORK5CYII=","orcid":"","institution":"Beijing Institute of Genomics","correspondingAuthor":true,"prefix":"","firstName":"Peilin","middleName":"","lastName":"Jia","suffix":""},{"id":602732218,"identity":"57b2acf0-614d-440a-95bd-f3f954d8f52c","order_by":1,"name":"Zhongming Zhao","email":"","orcid":"","institution":"Vanderbilt University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Zhongming","middleName":"","lastName":"Zhao","suffix":""}],"badges":[],"createdAt":"2026-01-19 03:39:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8634723/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8634723/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104780788,"identity":"3d8e05a5-0415-4468-9edc-1fcfb402a07f","added_by":"auto","created_at":"2026-03-17 07:53:58","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":269153,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSchematic overview of DrugSAGE.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-8634723/v1/88b05b9a7014887b59880ac1.png"},{"id":104472077,"identity":"1c9df082-f52b-4c43-914e-dfe58c4c21d1","added_by":"auto","created_at":"2026-03-12 07:29:01","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":305450,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDetermination of hyperparameters.\u003c/strong\u003e (A-C) Performance comparison of the combination strategy (A), the aggregation method for neighbor embeddings (B), and the activation function (C). (D-E) Investigation of the similarity of the most sensitive cell lines for drugs using gene co-expression. Each red line represents a drug. The boxplots represent the similarity of randomly selected cell lines 10,000 times. The number of cell lines to be compared is shown on the x-axis. (F) Model performance using the most correlated cell lines as neighbors to construct the network, or using random neighbors.\u003c/p\u003e","description":"","filename":"fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-8634723/v1/40992e9b9bc7e52f5ec7e083.png"},{"id":104472083,"identity":"5b956e16-3608-476e-af7a-62b876cafe20","added_by":"auto","created_at":"2026-03-12 07:29:01","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":518344,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eValidation using TCGA data.\u003c/strong\u003e (A) Comparison of predicted response to Lapatinib, a HER2+ inhibitor, using TCGA-BRCA samples stratified into negative, equivocal, and positive groups. Each dot is a TCGA-BRCA sample. (B) Comparison of predicted response to four MET inhibitors (PF2341066, PHA-665752, Crizotinib, and Foretinib) using TCGA-LUAD samples stratified based on the copy number variation status of the gene \u003cem\u003eMET\u003c/em\u003e. (C) Comparison of predicted response to EGFR inhibitors (Erlotinib, Gefitinib, Cetuximab, and Afatinib) using TCGA-LUAD samples stratified based on the EGFR mutation status. Note that Afatinib was tested under two different drug IDs in the GDSC1 dataset: Afatinib_1377 and Afatinib_1032. (D) Comparison of predicted responses to MEK1/2 inhibitors. In each cancer type, samples were stratified as mutant (e.g., KRAS mutation, HRAS mutant, NRAS mutant, or RAS mutant) and wild type (WT), and a one-sided t-test was applied to compare the two groups with the alternative hypothesis stated as the mutant group had a higher response than the WT group. For RAS genes, only mutations at the 12, 13, and 61 positions were considered. (E) Survival analysis results of TCGA-STAD samples stratified by the median value of their predicted response to 5-Fluorouracil. Red line: the high response group; blue line: the low response group. (F) Comparison of predicted responses to BRAF inhibitors. In each cancer type, samples were stratified as BRAF mutation versus WT, or BRAF or RAS mutation (BRAF|RAS) versus WT. A one-sided t-test was applied, similar to (D), where the p-values were labeled in each cell. For the row labels, T represents THCA, S for SKCM, and C for COAD.\u003c/p\u003e","description":"","filename":"fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-8634723/v1/a856a9f73d2d937ddfa38274.png"},{"id":104472079,"identity":"79883bcd-9812-411e-9a58-2ef12bc96b32","added_by":"auto","created_at":"2026-03-12 07:29:01","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":283116,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eValidation using independent datasets with bulk gene expression.\u003c/strong\u003e(A) Validation of models for Paclitaxel using three independent datasets. In each dataset, samples were classified as RD or pCR, and a t-test was applied to compare the predicted response by the corresponding model, i.e., CCLE or GDSC2. (B) Validation of the model for Cetuximab trained using GDSC1 reference data. Samples from GSE65021 were treated with a first line platinum and cetuximab and were classified into two groups, i.e., the long PFS group with 14 samples with a PFS exceeding 12 months and the short PFS group with 26 patients with a PFS less than 5.6 months. A t-test was applied to compared the predicted response. (C) Validation of the model for Cisplatin trained using the GDSC2 reference data. (D) Validation of the models for Erlotinib trained using the CCLE and GDSC2 reference data, respectively. E: epithelial-like; M: mesenchymal-like. (E) Validation of the models for Erlotinib trained using the CCLE with the default parameter R11N11 and GDSC2 (with the default parameter R6N6 and the drug-specific parameter R6N3) reference data, respectively. In each case, samples were stratified based on the median predicted response. (F) The distribution of Cox model z-scores from the models trained using CCLE R11N11, GDSC2 R6N6, and GDSC2 R6N3, as used in (E). (G) Validation of the models for PLX4720 using parental cell lines with matched resistant cell lines. (H) Validation models for PLX4720 using matched parental and resistant cell lines from the dataset GSE65185.\u003c/p\u003e","description":"","filename":"fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-8634723/v1/708f662e5d4263b1dbf64125.png"},{"id":104472082,"identity":"c84e0ce1-c50a-4c19-9d30-4746a548d063","added_by":"auto","created_at":"2026-03-12 07:29:01","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":139208,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eValidation using independent data with scRNA-seq data.\u003c/strong\u003e(A) Validation of models for the BET inhibitor, I-BET-762, by comparing parental cell lines with resistant cell lines from GSE110894. Two models were validated, the left using the direct models and the right using the model trained from scratch. (B) Validation of the models for Erlotinib in GSE149383 using resistant cell lines and sensitive cell lines. (C) Validation of models for Gefitinib using the direct and scratch models, respectively, to compare parental, early, and late cell lines. (D) Validation of the model for Bortezomib by comparing cell lines collected at four stages: t0, t12, t48, and t96. More details can be found in the main texts.\u003c/p\u003e","description":"","filename":"fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-8634723/v1/7ba44d8685ebfadbd83e0fe7.png"},{"id":104472080,"identity":"589377e3-2cf7-4349-8db8-d2e349879f35","added_by":"auto","created_at":"2026-03-12 07:29:01","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":190240,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparison of model performance.\u003c/strong\u003e Each panel illustrated the performance of six models for a drug from CCLE. The Y-axis is the R² in the test data.\u003c/p\u003e","description":"","filename":"fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-8634723/v1/b8705ee38b452faa9e6e09fb.png"},{"id":104784495,"identity":"dfa5eb62-e268-43b3-a802-f72ab4b20e0e","added_by":"auto","created_at":"2026-03-17 08:07:59","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2262541,"visible":true,"origin":"","legend":"","description":"","filename":"DrugSAGE.V3.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8634723/v1_covered_d6c2537e-08de-4c73-b95f-7e860581e2fe.pdf"},{"id":104780605,"identity":"cad65c7d-e348-4ea7-bd7f-5f0acc8f5cc9","added_by":"auto","created_at":"2026-03-17 07:53:21","extension":"xlsx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":23244,"visible":true,"origin":"","legend":"","description":"","filename":"TableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8634723/v1/433eab481b7f07f26b245670.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"DrugSAGE: an aggregation-based method for drug response imputation","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":"genome-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Genome Medicine](https://genomemedicine.biomedcentral.com/)","snPcode":"13073","submissionUrl":"https://submission.springernature.com/new-submission/13073/3","title":"Genome Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"GraphSAGE, Graph Neural Network, drug response","lastPublishedDoi":"10.21203/rs.3.rs-8634723/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8634723/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAccurate prediction of drug response is critical for optimizing the clinical application of cancer therapies. Nevertheless, owing to genetic and genomic heterogeneity and the complex tumor microenvironment, cancer samples may exhibit heterogeneous responses to the same therapeutic agents, even when harboring identical driver mutations or genes. In this study, we developed DrugSAGE, a novel framework that leverages Graph Neural Networks (GNNs) to model cellular heterogeneity and accurately predict drug response from transcriptomic data. DrugSAGE enhances prediction by aggregating features from a sample and its most similar counterparts. Critically, we introduce a customized linear layer incorporating gene-pathway annotations to provide biological interpretability, facilitating the identification of key pathways driving the prediction. We benchmarked DrugSAGE using TCGA bulk data, six public bulk datasets, and four scRNA-seq datasets. Predictions from DrugSAGE showed significant associations between drugs and their known target genes (e.g., HER2\u0026thinsp;+\u0026thinsp;inhibitors, MET inhibitors, BRAF inhibitors) or significant differences between patient groups stratified by their treatments (e.g., Erlotinib, PLX4720, 5-Fluorouracil). Furthermore, we show DrugSAGE can effectively predict drug response at the single-cell level. Additionally, by analyzing pathway-based embeddings, we identified pathways crucial to the models, including those involved in cancer, MAPK signaling, endocytosis, and JAK-STAT signaling, among others. Compared to previous methods, DrugSAGE demonstrated superior or comparable performance, offering a novel approach for predicting drug responses.\u003c/p\u003e","manuscriptTitle":"DrugSAGE: an aggregation-based method for drug response imputation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-12 07:28:49","doi":"10.21203/rs.3.rs-8634723/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-24T10:28:12+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-18T12:00:23+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-13T13:47:51+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-13T06:54:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"90579031815431982643705662144692868508","date":"2026-03-11T14:17:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"276917800083232074962133360133022603064","date":"2026-03-09T00:05:42+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-07T05:47:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"128608660267613727488018031164038352556","date":"2026-03-06T19:05:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"315929828850552331552381993051603422713","date":"2026-03-06T13:52:16+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-06T10:32:21+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-01-26T19:59:10+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-19T07:53:49+00:00","index":"","fulltext":""},{"type":"submitted","content":"Genome Medicine","date":"2026-01-19T03:30:51+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"genome-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Genome Medicine](https://genomemedicine.biomedcentral.com/)","snPcode":"13073","submissionUrl":"https://submission.springernature.com/new-submission/13073/3","title":"Genome Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d328f629-a183-4b2f-9873-7865741020f1","owner":[],"postedDate":"March 12th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-19T04:08:27+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-12 07:28:49","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8634723","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8634723","identity":"rs-8634723","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}