{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,31]],"date-time":"2026-03-31T16:46:54Z","timestamp":1774975614393,"version":"3.50.1"},"reference-count":59,"publisher":"Oxford University Press (OUP)","issue":"Supplement_1","license":[{"start":{"date-parts":[[2024,6,28]],"date-time":"2024-06-28T00:00:00Z","timestamp":1719532800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/100000002","name":"National Institutes of Health","doi-asserted-by":"publisher","award":["DA051906-04"],"award-info":[{"award-number":["DA051906-04"]}],"id":[{"id":"10.13039\/100000002","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2024,6,28]]},"abstract":"<jats:title>Abstract<\/jats:title>\n               <jats:sec>\n                  <jats:title>Motivation<\/jats:title>\n                  <jats:p>The current paradigm of deep learning models for the joint representation of molecules and text primarily relies on 1D or 2D molecular formats, neglecting significant 3D structural information that offers valuable physical insight. This narrow focus inhibits the models\u2019 versatility and adaptability across a wide range of modalities. Conversely, the limited research focusing on explicit 3D representation tends to overlook textual data within the biomedical domain.<\/jats:p>\n               <\/jats:sec>\n               <jats:sec>\n                  <jats:title>Results<\/jats:title>\n                  <jats:p>We present a unified pre-trained language model, MolLM, that concurrently captures 2D and 3D molecular information alongside biomedical text. MolLM consists of a text Transformer encoder and a molecular Transformer encoder, designed to encode both 2D and 3D molecular structures. To support MolLM\u2019s self-supervised pre-training, we constructed 160K molecule-text pairings. Employing contrastive learning as a supervisory signal for learning, MolLM demonstrates robust molecular representation capabilities across four downstream tasks, including cross-modal molecule and text matching, property prediction, captioning, and text-prompted molecular editing. Through ablation, we demonstrate that the inclusion of explicit 3D representations improves performance in these downstream tasks.<\/jats:p>\n               <\/jats:sec>\n               <jats:sec>\n                  <jats:title>Availability and implementation<\/jats:title>\n                  <jats:p>Our code, data, pre-trained model weights, and examples of using our model are all available at https:\/\/github.com\/gersteinlab\/MolLM. In particular, we provide Jupyter Notebooks offering step-by-step guidance on how to use MolLM to extract embeddings for both molecules and text.<\/jats:p>\n               <\/jats:sec>","DOI":"10.1093\/bioinformatics\/btae260","type":"journal-article","created":{"date-parts":[[2024,6,28]],"date-time":"2024-06-28T09:30:32Z","timestamp":1719567032000},"page":"i357-i368","source":"Crossref","is-referenced-by-count":13,"title":["<i>MolLM<\/i>: a unified language model for integrating biomedical text with 2D and 3D molecular representations"],"prefix":"10.1093","volume":"40","author":[{"ORCID":"https:\/\/orcid.org\/0009-0006-2700-4513","authenticated-orcid":false,"given":"Xiangru","family":"Tang","sequence":"first","affiliation":[{"name":"Department of Computer Science, Yale University, New Haven, CT 06520, United States"}]},{"given":"Andrew","family":"Tran","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Yale University, New Haven, CT 06520, United States"}]},{"given":"Jeffrey","family":"Tan","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Yale University, New Haven, CT 06520, United States"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-9746-3719","authenticated-orcid":false,"given":"Mark B","family":"Gerstein","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Yale University, New Haven, CT 06520, United States"},{"name":"Program in Computational Biology & Bioinformatics, Yale University, New Haven, CT 06520, United States"},{"name":"Department of Molecular Biophysics & Biochemistry, Yale University, New Haven, CT 06520, United States"},{"name":"Department of Statistics & Data Science, Yale University, New Haven, CT 06520, United States"},{"name":"Department of Biomedical Informatics & Data Science, Yale University, New Haven, CT 06520, United States"}]}],"member":"286","published-online":{"date-parts":[[2024,6,28]]},"reference":[{"key":"2024071814163753100_btae260-B1","author":"An"},{"key":"2024071814163753100_btae260-B2","doi-asserted-by":"crossref","first-page":"90","DOI":"10.1038\/nchem.1243","article-title":"Quantifying the chemical beauty of drugs","volume":"4","author":"Bickerton","year":"2012","journal-title":"Nat Chem"},{"key":"2024071814163753100_btae260-B3","doi-asserted-by":"crossref","first-page":"232","DOI":"10.1093\/bib\/bbz164","article-title":"A survey and systematic assessment of computational methods for drug response prediction","volume":"22","author":"Chen","year":"2021","journal-title":"Brief Bioinform"},{"key":"2024071814163753100_btae260-B4","author":"Chilingaryan"},{"key":"2024071814163753100_btae260-B5","doi-asserted-by":"crossref","first-page":"1757","DOI":"10.1021\/acs.jcim.6b00601","article-title":"Convolutional embedding of attributed molecular graphs for physical property prediction","volume":"57","author":"Coley","year":"2017","journal-title":"J Chem Inf Model"},{"key":"2024071814163753100_btae260-B6","doi-asserted-by":"crossref","first-page":"194","DOI":"10.1016\/j.jmgm.2014.10.006","article-title":"3d-morse descriptors explained","volume":"54","author":"Devinyak","year":"2014","journal-title":"J Mol Graph Model"},{"key":"2024071814163753100_btae260-B7","author":"Devlin"},{"key":"2024071814163753100_btae260-B8","first-page":"375","author":"Edwards","year":"2022"},{"key":"2024071814163753100_btae260-B9","doi-asserted-by":"crossref","first-page":"542","DOI":"10.1038\/s42256-023-00654-0","article-title":"Knowledge graph-enhanced molecular contrastive learning with functional prompt","volume":"5","author":"Fang","year":"2023","journal-title":"Nat Mach Intell"},{"key":"2024071814163753100_btae260-B10","doi-asserted-by":"crossref","first-page":"1","DOI":"10.1145\/3458754","article-title":"Domain-specific language model pretraining for biomedical natural language processing","volume":"3","author":"Gu","year":"2021","journal-title":"ACM Trans Comput Healthcare"},{"key":"2024071814163753100_btae260-B11","first-page":"22118","article-title":"Open graph benchmark: datasets for machine learning on graphs","volume":"33","author":"Hu","year":"2020","journal-title":"Adv Neural Inf Process Syst."},{"key":"2024071814163753100_btae260-B12","doi-asserted-by":"crossref","first-page":"5936","DOI":"10.1021\/acs.jcim.0c00416","article-title":"Comprehensive study on molecular supervised learning with graph neural networks","volume":"60","author":"Hwang","year":"2020","journal-title":"J Chem Inf Model"},{"key":"2024071814163753100_btae260-B13","doi-asserted-by":"crossref","first-page":"1691","DOI":"10.1021\/acs.jcim.0c01294","article-title":"Ggl-tox: geometric graph learning for toxicity prediction","volume":"61","author":"Jiang","year":"2021","journal-title":"J Chem Inf Model"},{"key":"2024071814163753100_btae260-B14","doi-asserted-by":"crossref","first-page":"16991","DOI":"10.1038\/s41598-017-17299-w","article-title":"A universal 3d voxel descriptor for solid-state material informatics with deep convolutional neural networks","volume":"7","author":"Kajita","year":"2017","journal-title":"Sci Rep"},{"key":"2024071814163753100_btae260-B15","author":"Kim"},{"key":"2024071814163753100_btae260-B16","author":"Kipf"},{"key":"2024071814163753100_btae260-B17","doi-asserted-by":"crossref","first-page":"672","DOI":"10.1016\/j.ccell.2020.09.014","article-title":"Predicting drug response and synergy using a deep learning model of human cancer cells","volume":"38","author":"Kuenzi","year":"2020","journal-title":"Cancer Cell"},{"key":"2024071814163753100_btae260-B18","doi-asserted-by":"crossref","first-page":"4378","DOI":"10.1021\/acs.molpharmaceut.7b01134","article-title":"3d molecular representations based on the wave transform for convolutional neural networks","volume":"15","author":"Kuzminykh","year":"2018","journal-title":"Mol Pharm"},{"key":"2024071814163753100_btae260-B19","author":"Landrum"},{"key":"2024071814163753100_btae260-B20","author":"Li"},{"key":"2024071814163753100_btae260-B21","author":"Li"},{"key":"2024071814163753100_btae260-B22","author":"Li"},{"key":"2024071814163753100_btae260-B23","author":"Liu"},{"key":"2024071814163753100_btae260-B24","doi-asserted-by":"crossref","first-page":"106524","DOI":"10.1016\/j.compbiomed.2022.106524","article-title":"The prediction of molecular toxicity based on bigru and graphsage","volume":"153","author":"Liu","year":"2023","journal-title":"Comput Biol Med"},{"key":"2024071814163753100_btae260-B25","doi-asserted-by":"crossref","first-page":"1447","DOI":"10.1038\/s42256-023-00759-6","article-title":"Multi-modal molecule structure-text model for text-based retrieval and editing","volume":"5","author":"Liu","year":"2023","journal-title":"Nat Mach Intell"},{"key":"2024071814163753100_btae260-B26","first-page":"4969","author":"Lo","year":"2020"},{"key":"2024071814163753100_btae260-B27","doi-asserted-by":"crossref","first-page":"bbac409","DOI":"10.1093\/bib\/bbac409","article-title":"Biogpt: generative pre-trained transformer for biomedical text generation and mining","volume":"23","author":"Luo","year":"2022","journal-title":"Brief Bioinform"},{"key":"2024071814163753100_btae260-B28","author":"Luo"},{"key":"2024071814163753100_btae260-B29","doi-asserted-by":"crossref","first-page":"1206","DOI":"10.1021\/acs.chemrestox.2c00384","article-title":"Recent advances in toxicity prediction: applications of deep graph learning","volume":"36","author":"Miao","year":"2023","journal-title":"Chem Res Toxicol"},{"key":"2024071814163753100_btae260-B30","author":"Pei"},{"key":"2024071814163753100_btae260-B31","doi-asserted-by":"crossref","first-page":"1136","DOI":"10.1021\/acs.jcim.8b00626","article-title":"De novo molecule design by translating from reduced graphs to smiles","volume":"59","author":"Pog\u00e1ny","year":"2018","journal-title":"J Chem Inf Model"},{"key":"2024071814163753100_btae260-B32","author":"Radford","year":"2018"},{"key":"2024071814163753100_btae260-B33","first-page":"1","article-title":"Exploring the limits of transfer learning with a unified text-to-text transformer","volume":"21","author":"Raffel","year":"2020","journal-title":"J Mach Learn Res."},{"key":"2024071814163753100_btae260-B34","first-page":"12559","article-title":"Self-supervised graph transformer on large-scale molecular data","volume":"33","author":"Rong","year":"2020","journal-title":"Adv Neural Inf Process Syst."},{"key":"2024071814163753100_btae260-B35","doi-asserted-by":"crossref","first-page":"1256","DOI":"10.1038\/s42256-022-00580-7","article-title":"Large-scale chemical language representations capture molecular structure and properties","volume":"4","author":"Ross","year":"2022","journal-title":"Nat Mach Intell."},{"key":"2024071814163753100_btae260-B36","doi-asserted-by":"crossref","first-page":"2758","DOI":"10.1109\/78.650102","article-title":"Comparing support vector machines with gaussian kernels to radial basis function classifiers","volume":"45","author":"Scholkopf","year":"1997","journal-title":"IEEE Trans Signal Process"},{"key":"2024071814163753100_btae260-B37","first-page":"172","author":"Singhal"},{"key":"2024071814163753100_btae260-B38","author":"St\u00e4rk"},{"key":"2024071814163753100_btae260-B39","doi-asserted-by":"crossref","first-page":"2324","DOI":"10.1021\/acs.jcim.5b00559","article-title":"Zinc 15\u2014ligand discovery for everyone","volume":"55","author":"Sterling","year":"2015","journal-title":"J Chem Inf Model"},{"key":"2024071814163753100_btae260-B40","author":"Su"},{"key":"2024071814163753100_btae260-B41","author":"Thomas"},{"key":"2024071814163753100_btae260-B42","author":"Vaswani"},{"key":"2024071814163753100_btae260-B43","first-page":"429","author":"Wang","year":"2019"},{"key":"2024071814163753100_btae260-B44","doi-asserted-by":"crossref","first-page":"2713","DOI":"10.1021\/acs.jcim.2c00495","article-title":"Improving molecular contrastive learning via faulty negative mitigation and decomposed fragment contrast","volume":"62","author":"Wang","year":"2022","journal-title":"J Chem Inf Model"},{"key":"2024071814163753100_btae260-B45","doi-asserted-by":"crossref","first-page":"279","DOI":"10.1038\/s42256-022-00447-x","article-title":"Molecular contrastive learning of representations via graph neural networks","volume":"4","author":"Wang","year":"2022","journal-title":"Nat Mach Intell"},{"key":"2024071814163753100_btae260-B46","author":"Wei"},{"key":"2024071814163753100_btae260-B47","doi-asserted-by":"crossref","first-page":"31","DOI":"10.1021\/ci00057a005","article-title":"Smiles, a chemical language and information system. 1. Introduction to methodology and encoding rules","volume":"28","author":"Weininger","year":"1988","journal-title":"J Chem Inf Comput Sci"},{"key":"2024071814163753100_btae260-B48","doi-asserted-by":"crossref","first-page":"868","DOI":"10.1021\/ci990307l","article-title":"Prediction of physicochemical parameters by atomic contributions","volume":"39","author":"Wildman","year":"1999","journal-title":"J Chem Inf Comput Sci"},{"key":"2024071814163753100_btae260-B49","first-page":"513","author":"Wu"},{"key":"2024071814163753100_btae260-B50","author":"Xia"},{"key":"2024071814163753100_btae260-B51","author":"Xu"},{"key":"2024071814163753100_btae260-B52","doi-asserted-by":"crossref","first-page":"3370","DOI":"10.1021\/acs.jcim.9b00237","article-title":"Analyzing learned molecular representations for property prediction","volume":"59","author":"Yang","year":"2019","journal-title":"J Chem Inf Model."},{"key":"2024071814163753100_btae260-B53","doi-asserted-by":"crossref","first-page":"323","DOI":"10.1093\/bfgp\/elab036","article-title":"Review of unsupervised pretraining strategies for molecules representation","volume":"20","author":"Yu","year":"2021","journal-title":"Brief Funct Genomics"},{"key":"2024071814163753100_btae260-B54","author":"Zang"},{"key":"2024071814163753100_btae260-B55","doi-asserted-by":"crossref","first-page":"862","DOI":"10.1038\/s41467-022-28494-3","article-title":"A deep-learning system bridging molecule structure and biomedical text with comprehension comparable to human professionals","volume":"13","author":"Zeng","year":"2022","journal-title":"Nat Commun"},{"key":"2024071814163753100_btae260-B56","first-page":"15870","article-title":"Motif-based graph self-supervised learning for molecular property prediction","volume":"34","author":"Zhang","year":"2021","journal-title":"Adv Neural Inf Process Syst"},{"key":"2024071814163753100_btae260-B57","author":"Zhu"},{"key":"2024071814163753100_btae260-B58","author":"Zhu"},{"key":"2024071814163753100_btae260-B59","doi-asserted-by":"crossref","first-page":"1856","DOI":"10.1021\/acs.accounts.0c00770","article-title":"Predicting reaction yields via supervised learning","volume":"54","author":"Zuranski","year":"2021","journal-title":"Acc Chem Res."}],"container-title":["Bioinformatics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article-pdf\/40\/Supplement_1\/i357\/58585840\/btae260.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article-pdf\/40\/Supplement_1\/i357\/58585840\/btae260.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2024,7,18]],"date-time":"2024-07-18T15:36:40Z","timestamp":1721317000000},"score":1,"resource":{"primary":{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article\/40\/Supplement_1\/i357\/7700902"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,6,28]]},"references-count":59,"journal-issue":{"issue":"Supplement_1","published-print":{"date-parts":[[2024,6,28]]}},"URL":"https:\/\/doi.org\/10.1093\/bioinformatics\/btae260","relation":{},"ISSN":["1367-4803","1367-4811"],"issn-type":[{"value":"1367-4803","type":"print"},{"value":"1367-4811","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2024,7]]},"published":{"date-parts":[[2024,6,28]]}}}