{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/deep-learning-to-generate-in-silico-chemical","title":"Deep learning to generate in silico chemical property libraries and candidate molecules for small molecule identification in complex samples","arxiv_id":"1905.08411","date":"2019-05-21","proceeding":null,"authors":[],"abstract":"Comprehensive and unambiguous identification of small molecules in complex\nsamples will revolutionize our understanding of the role of metabolites in\nbiological systems. Existing and emerging technologies have enabled measurement\nof chemical properties of molecules in complex mixtures and, in concert, are\nsensitive enough to resolve even stereoisomers. Despite these experimental\nadvances, small molecule identification is inhibited by (i) chemical reference\nlibraries representing <1% of known molecules, limiting the number of possible\nidentifications, and (ii) the lack of a method to generate candidate matches\ndirectly from experimental features (i.e. without a library). To this end, we\ndeveloped a variational autoencoder (VAE) to learn a continuous numerical, or\nlatent, representation of molecular structure to expand reference libraries for\nsmall molecule identification. We extended the VAE to include a chemical\nproperty decoder, trained as a multitask network, in order to shape the latent\nrepresentation such that it assembles according to desired chemical properties.\nThe approach is unique in its application to small molecule identification,\nwith its focus on m/z and CCS, paired with its training paradigm, which\ninvolved a cascade of transfer learning iterations. This allows the network to\nlearn as much as possible at each stage, enabling success with progressively\nsmaller datasets without overfitting. Once trained, the network can rapidly\npredict chemical properties directly from structure, as well as generate\ncandidate structures with desired chemical properties. Additionally, the\nability to generate novel molecules along manifolds, defined by chemical\nproperty analogues, positions DarkChem as highly useful in a number of\napplication areas, including metabolomics and small molecule identification,\ndrug discovery and design, chemical forensics, and beyond.","url_abs":"http://arxiv.org/abs/1905.08411v1","url_pdf":"http://arxiv.org/pdf/1905.08411v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"deep-learning-to-generate-in-silico-chemical","repo_url":"https://github.com/pnnl/darkchem","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"drug-discovery","task_name":"Drug Discovery"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}