A Multi-Task Hybrid GNN–Transformer Framework for Simultaneous Product SMILES Generation and Reaction Type Classification

A Multi-Task Hybrid GNN–Transformer Framework for Simultaneous Product SMILES Generation and Reaction Type Classification | 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 Research Article A Multi-Task Hybrid GNN–Transformer Framework for Simultaneous Product SMILES Generation and Reaction Type Classification Doaa Sayed, Mohammed M. Abbassy, Mohammed Abdalla This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9534356/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 Despite their basic connection, predicting reaction products and categorizing reaction mechanisms are typically handled as distinct tasks in computational chemistry. We describe a unified multi-task GNN--Transformer architecture that uses a shared molecular graph representation to simultaneously generate SMILES and classify reaction types. The model enables fine-grained alignment between molecular structure and produced sequences by combining an autoregressive Transformer decoder augmented by atom-level cross-attention with a GINEConv-based encoder. Reaction types are predicted by a classification head, and task balance is managed by a progressive training schedule and learnt uncertainty weighting. Learning rate and label smoothing are the main factors influencing generation quality, accounting for 60% of BLEU variance, according to a systematic hyperparameter optimization study (30 Optuna trials). Across ten reaction classes, the suggested multi-task model achieves 88.91% token accuracy, 99.71% Top-5 accuracy, BLEU = 0.497, and 38.93% classification accuracy. Significant improvements are obtained by scaling the model to a bigger setting (v4-LARGE) with 142k training samples and curriculum learning over increasing 1 sequence lengths, attaining 95.37% token accuracy and BLEU = 0.548. Atomlevel cross-attention, class balancing, and SMILES augmentation are important performance factors, as confirmed by ablation analysis. These findings show that scalable and efficient multi-task molecular modeling is made possible by combining atom-level attention, adaptive loss weighting, and curriculum learning, creating a solid foundation for reaction prediction. graph neural networks transformer decoder SMILES generation reaction type classification multi-task learning drug discovery Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 10 May, 2026 Reviewers agreed at journal 07 May, 2026 Reviewers agreed at journal 07 May, 2026 Reviewers agreed at journal 05 May, 2026 Reviewers invited by journal 05 May, 2026 Editor assigned by journal 01 May, 2026 Submission checks completed at journal 29 Apr, 2026 First submitted to journal 26 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-9534356","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":637809441,"identity":"5dd8a5b2-8934-475b-ab24-468219d006d0","order_by":0,"name":"Doaa Sayed","email":"data:image/png;base64,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","orcid":"","institution":"Faculty of Computers and Artificial Intelligence, Beni-Suef University","correspondingAuthor":true,"prefix":"","firstName":"Doaa","middleName":"","lastName":"Sayed","suffix":""},{"id":637809442,"identity":"9c7d37d0-0b41-4984-8936-85093044a7fe","order_by":1,"name":"Mohammed M. Abbassy","email":"","orcid":"","institution":"Faculty of Computers and Artificial Intelligence, Beni-Suef University","correspondingAuthor":false,"prefix":"","firstName":"Mohammed","middleName":"M.","lastName":"Abbassy","suffix":""},{"id":637809443,"identity":"06d7b428-6b3b-4d77-8aea-1d88d1ebbdf4","order_by":2,"name":"Mohammed Abdalla","email":"","orcid":"","institution":"Faculty of Computers and Artificial Intelligence, Beni-Suef University","correspondingAuthor":false,"prefix":"","firstName":"Mohammed","middleName":"","lastName":"Abdalla","suffix":""}],"badges":[],"createdAt":"2026-04-26 20:23:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9534356/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9534356/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109204824,"identity":"f55ab5cb-78ba-4332-9f03-73cb3e00e8b1","added_by":"auto","created_at":"2026-05-13 15:02:25","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1874098,"visible":true,"origin":"","legend":"","description":"","filename":"Multitask2.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9534356/v1_covered_ffe02e50-eae1-47f5-9342-0ff25a98ade7.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Multi-Task Hybrid GNN–Transformer Framework for Simultaneous Product SMILES Generation and Reaction Type Classification","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":"international-journal-of-computational-intelligence-systems","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [International Journal of Computational Intelligence Systems](https://link.springer.com/journal/44196)","snPcode":"44196","submissionUrl":"https://submission.springernature.com/new-submission/44196/3","title":"International Journal of Computational Intelligence Systems","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"graph neural networks, transformer decoder, SMILES generation, reaction type classification, multi-task learning, drug discovery","lastPublishedDoi":"10.21203/rs.3.rs-9534356/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9534356/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDespite their basic connection, predicting reaction products and categorizing reaction mechanisms are typically handled as distinct tasks in computational chemistry. We describe a unified multi-task GNN--Transformer architecture that uses a shared molecular graph representation to simultaneously generate SMILES and classify reaction types. The model enables fine-grained alignment between molecular structure and produced sequences by combining an autoregressive Transformer decoder augmented by atom-level cross-attention with a GINEConv-based encoder. Reaction types are predicted by a classification head, and task balance is managed by a progressive training schedule and learnt uncertainty weighting.\u003c/p\u003e\n\u003cp\u003eLearning rate and label smoothing are the main factors influencing generation quality, accounting for 60% of BLEU variance, according to a systematic hyperparameter optimization study (30 Optuna trials). Across ten reaction classes, the suggested multi-task model achieves 88.91% token accuracy, 99.71% Top-5 accuracy, BLEU = 0.497, and 38.93% classification accuracy.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSignificant improvements are obtained by scaling the model to a bigger setting (v4-LARGE) with 142k training samples and curriculum learning over increasing 1 sequence lengths, attaining 95.37% token accuracy and BLEU = 0.548. Atomlevel cross-attention, class balancing, and SMILES augmentation are important performance factors, as confirmed by ablation analysis.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThese findings show that scalable and efficient multi-task molecular modeling is made possible by combining atom-level attention, adaptive loss weighting, and curriculum learning, creating a solid foundation for reaction prediction.\u003c/p\u003e","manuscriptTitle":"A Multi-Task Hybrid GNN–Transformer Framework for Simultaneous Product SMILES Generation and Reaction Type Classification","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-12 14:52:57","doi":"10.21203/rs.3.rs-9534356/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"122728917489294789284467592280345799608","date":"2026-05-11T00:29:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"150543868760839936008759287723928837341","date":"2026-05-07T23:55:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"162263580432511420162150502308848195502","date":"2026-05-07T10:45:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"125320980565139519256810417270118252667","date":"2026-05-05T12:26:14+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-05-05T09:53:20+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-01T05:55:53+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-30T02:31:31+00:00","index":"","fulltext":""},{"type":"submitted","content":"International Journal of Computational Intelligence Systems","date":"2026-04-26T20:10:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-computational-intelligence-systems","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [International Journal of Computational Intelligence Systems](https://link.springer.com/journal/44196)","snPcode":"44196","submissionUrl":"https://submission.springernature.com/new-submission/44196/3","title":"International Journal of Computational Intelligence Systems","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"66851a40-6f94-41f4-83c8-721f6862c9fb","owner":[],"postedDate":"May 12th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"122728917489294789284467592280345799608","date":"2026-05-11T00:29:31+00:00","index":27,"fulltext":""},{"type":"reviewerAgreed","content":"150543868760839936008759287723928837341","date":"2026-05-07T23:55:54+00:00","index":25,"fulltext":""},{"type":"reviewerAgreed","content":"162263580432511420162150502308848195502","date":"2026-05-07T10:45:33+00:00","index":24,"fulltext":""},{"type":"reviewerAgreed","content":"125320980565139519256810417270118252667","date":"2026-05-05T12:26:14+00:00","index":22,"fulltext":""},{"type":"reviewersInvited","content":"15","date":"2026-05-05T09:53:20+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-01T05:55:53+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-30T02:31:31+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-12T14:52:58+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-12 14:52:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9534356","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9534356","identity":"rs-9534356","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}