LS-MolGen: Ligand-and-Structure Dual-driven Deep Reinforcement Learning for Target-specific Molecular Generation Improves Binding Affinity and Novelty

LS-MolGen: Ligand-and-Structure Dual-driven Deep Reinforcement Learning for Target-specific Molecular Generation Improves Binding Affinity and Novelty | 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 LS-MolGen: Ligand-and-Structure Dual-driven Deep Reinforcement Learning for Target-specific Molecular Generation Improves Binding Affinity and Novelty Song Li, Chao Hu, Song Ke, Chenxing Yang, Jun Chen, Yi Xiong, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2793302/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Molecule generative models based on deep learning have attracted significant attention in de novo drug design. However, most current generative approaches are either only ligand-based or only structure-based, which do not leverage the complementary knowledge from ligands and the structure of binding target. In this work, we proposed a new ligand and structure combined molecular generative model, LS-MolGen, that integrates representation learning, transfer learning, and reinforcement learning. Focus knowledge from transfer learning and special explore strategy in reinforcement learning enables LS-MolGen to generate novel and active molecules efficiently. The results of evaluation using EGFR and case study of inhibitor design for SARS-CoV-2 Mpro showed that LS-MolGen outperformed other state-of-the-art ligand-based or structure-based generative models and was capable of de novo designing promising compounds with novel scaffold and high binding affinity. Thus, we recommend that this proof-of-concept ligand-and-structure-based generative model will provide a promising new tool for target-specific molecular generation and drug design. molecule generation transfer learning reinforcement learning ligand-and-structure-based drug design Figures Figure 1 Figure 2 Figure 3 Figure 4 Full Text Additional Declarations Competing interest reported. Song Ke, Chenxing Yang, and Jun Chen are employees of Shanghai Matwings Technology Co., Ltd., Shanghai. Other authors declare no competing interests. Supplementary Files supportinginformation.pdf Cite Share Download PDF Status: Posted Version 1 posted 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-2793302","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Method Article","associatedPublications":[],"authors":[{"id":190482068,"identity":"b0ab040f-2cbe-4900-9736-17cd84240c4c","order_by":0,"name":"Song Li","email":"","orcid":"","institution":"Shanghai Jiao Tong University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Song","middleName":"","lastName":"Li","suffix":""},{"id":190482069,"identity":"746ba5f1-0d2d-4985-8fea-1c66431014e5","order_by":1,"name":"Chao Hu","email":"","orcid":"","institution":"East China University of Science and Technology","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Chao","middleName":"","lastName":"Hu","suffix":""},{"id":190482070,"identity":"e1910e25-00c4-4ee2-9640-f05fe711f1ec","order_by":2,"name":"Song Ke","email":"","orcid":"","institution":"Shanghai Matwings Technology Co., Ltd.","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Song","middleName":"","lastName":"Ke","suffix":""},{"id":190482071,"identity":"831656fd-49ae-4895-a24a-d34e66c16c9b","order_by":3,"name":"Chenxing Yang","email":"","orcid":"","institution":"Shanghai Matwings Technology Co., Ltd.","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Chenxing","middleName":"","lastName":"Yang","suffix":""},{"id":190482072,"identity":"a7c9dc10-8097-43a1-a689-fd9d610b7414","order_by":4,"name":"Jun Chen","email":"","orcid":"","institution":"Shanghai Matwings Technology Co., Ltd.","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Jun","middleName":"","lastName":"Chen","suffix":""},{"id":190482073,"identity":"e20ec1c1-e523-427d-b102-8735532b8967","order_by":5,"name":"Yi Xiong","email":"","orcid":"","institution":"Shanghai Jiao Tong University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Yi","middleName":"","lastName":"Xiong","suffix":""},{"id":190482074,"identity":"fc92e69c-bf0d-49f0-9b9c-b95d582b93f0","order_by":6,"name":"Hao Liu","email":"","orcid":"","institution":"Shanghai Jiao Tong University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Hao","middleName":"","lastName":"Liu","suffix":""},{"id":190482075,"identity":"95583566-ad5d-44bf-9ce8-ed9baac4fec6","order_by":7,"name":"Liang Hong","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAElEQVRIiWNgGAWjYBACAxDx4IcFD5BifMDAgxDEryWxRwKkltmAgceASC0JbBIgCkQSocWcvffwiwQeCRn+2e3XKn/I/ElsYG/eJsFQcwenFsuec2kWCRYSPBJ3zpTd5uExSGzgOVYmwXDsGW6H3cgxMwDawsNwIyftNgNIi0SOmQRjw2ECWtgkeOSBWgp/gLTIvyGoxfgBSIvBjfRjDGCHSfAQ0HLmjBk4kA1v5DBL8/AYG7fxpBVbJBzDo+V4j/GHDz9s7OVupD/8+LNHTraf/fDGGx9qcGthgEQHCACjkbEHyAWxE/BpAEb6BwjN/oCB4Qd+paNgFIyCUTAyAQD2aFFLbDSbGQAAAABJRU5ErkJggg==","orcid":"","institution":"Shanghai Jiao Tong University","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Liang","middleName":"","lastName":"Hong","suffix":""}],"badges":[],"createdAt":"2023-04-08 14:59:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2793302/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2793302/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":35646871,"identity":"abcb60b9-87a9-444c-9394-0a9a2a73af91","added_by":"auto","created_at":"2023-04-12 13:25:49","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1223000,"visible":true,"origin":"","legend":"\u003cp\u003eThe concept of molecular generative models, and the architecture of LS-MolGen approach. (a) The concept of ligand-based method, (b) structure-based method, and (c) ligand-and-structure-based method. Illustration in a, b, and c, the green region represents the available chemical space, the yellow region represents the chemical space of known bioactive ligands, and the blue arrow represents the exploration of chemical space. (d) The pipeline of the LS-MolGen approach. (e) The architecture of the prior model with RNN. (f) The illustration of transfer learning model. (g) Exploration of chemical space of reinforcement learning combined with molecular docking.\u003c/p\u003e","description":"","filename":"fig1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-2793302/v1/067319e2786d88cd239fd549.jpg"},{"id":35647611,"identity":"439e59d9-9ee2-4390-bdb1-4d02b8ab3c5c","added_by":"auto","created_at":"2023-04-12 13:33:49","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":529907,"visible":true,"origin":"","legend":"\u003cp\u003eQuantitative evaluations of LS-MolGen. (a) Evolution of the explored molecular average score in the reinforcement learning loop. (b) Chemical space of generated molecules and bioactive ligands of EGFR visualized by t-SNE dimensionality reduction. (c) Distribution of molecular docking score to EGFR. (d-f) Distributions of properties, including molecular weight, QED, and SA score.\u003c/p\u003e","description":"","filename":"fig2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-2793302/v1/fb59f171503a795a662bf162.jpg"},{"id":35646868,"identity":"d231a9f2-e93b-412b-9cf9-2ed2ce428a15","added_by":"auto","created_at":"2023-04-12 13:25:49","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":523399,"visible":true,"origin":"","legend":"\u003cp\u003eThree example inhibitors of Mpro and generated compounds that match their pharmacophores. The first column is the docked pose and pharmacophoric groups of the inhibitor, the second and third columns are 2D molecular structure of the inhibitor and the matched generated compound respectively (scaffold highlighted), the fourth column is the molecular pharmacophores supposition, and the last column is the RMSD of supposition and similarity between inhibitor and generated compound.\u003c/p\u003e","description":"","filename":"fig3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-2793302/v1/a0b643b17621504e3adcdea5.jpg"},{"id":35646872,"identity":"09306b9b-507f-4912-b2ff-324735eaaebf","added_by":"auto","created_at":"2023-04-12 13:25:49","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":326261,"visible":true,"origin":"","legend":"\u003cp\u003eDocked poses of generated compounds. Hydrogen bond was represented as a red dashed line. Docking protocols were performed by LeDock.\u003c/p\u003e","description":"","filename":"fig4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-2793302/v1/c2a5b7a9caf81800d760895e.jpg"},{"id":35647612,"identity":"dc500464-013f-40ab-83ad-842bd4be1eeb","added_by":"auto","created_at":"2023-04-12 13:34:02","extension":"pdf","order_by":6,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":811940,"visible":true,"origin":"","legend":"","description":"","filename":"LSMolGenBMC.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2793302/v1_covered.pdf"},{"id":35646870,"identity":"76e32929-f891-4f0f-9e2f-e506424482ca","added_by":"auto","created_at":"2023-04-12 13:25:49","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":891689,"visible":true,"origin":"","legend":"","description":"","filename":"supportinginformation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2793302/v1/8418570269ae58e462a3fc4a.pdf"}],"financialInterests":"Competing interest reported. Song Ke, Chenxing Yang, and Jun Chen are employees of Shanghai Matwings Technology Co., Ltd., Shanghai. Other authors declare no competing interests.","formattedTitle":"LS-MolGen: Ligand-and-Structure Dual-driven Deep Reinforcement Learning for Target-specific Molecular Generation Improves Binding Affinity and Novelty","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"molecule generation, transfer learning, reinforcement learning, ligand-and-structure-based drug design","lastPublishedDoi":"10.21203/rs.3.rs-2793302/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2793302/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Molecule generative models based on deep learning have attracted significant attention in de novo drug design. However, most current generative approaches are either only ligand-based or only structure-based, which do not leverage the complementary knowledge from ligands and the structure of binding target. In this work, we proposed a new ligand and structure combined molecular generative model, LS-MolGen, that integrates representation learning, transfer learning, and reinforcement learning. Focus knowledge from transfer learning and special explore strategy in reinforcement learning enables LS-MolGen to generate novel and active molecules efficiently. The results of evaluation using EGFR and case study of inhibitor design for SARS-CoV-2 Mpro showed that LS-MolGen outperformed other state-of-the-art ligand-based or structure-based generative models and was capable of de novo designing promising compounds with novel scaffold and high binding affinity. Thus, we recommend that this proof-of-concept ligand-and-structure-based generative model will provide a promising new tool for target-specific molecular generation and drug design.","manuscriptTitle":"LS-MolGen: Ligand-and-Structure Dual-driven Deep Reinforcement Learning for Target-specific Molecular Generation Improves Binding Affinity and Novelty","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-04-12 13:25:45","doi":"10.21203/rs.3.rs-2793302/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"cfc2f9ed-6530-4927-a077-3b3052d71a78","owner":[],"postedDate":"April 12th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2023-04-17T12:14:31+00:00","versionOfRecord":[],"versionCreatedAt":"2023-04-12 13:25:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-2793302","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2793302","identity":"rs-2793302","version":["v1"]},"buildId":"cBFmMYwuxLRRLfASyISRj","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}