{"title":"Predicting 3D RNA structure from the nucleotide sequence using Euclidean neural networks.","authors":"Congzhou M Sha, Jian Wang, Nikolay V Dokholyan","doi":"10.1016/j.bpj.2023.10.011","DOIUrl":null,"url":null,"abstract":"<p><p>Fast and accurate 3D RNA structure prediction remains a major challenge in structural biology, mostly due to the size and flexibility of RNA molecules, as well as the lack of diverse experimentally determined structures of RNA molecules. Unlike DNA structure, RNA structure is far less constrained by basepair hydrogen bonding, resulting in an explosion of potential stable states. Here, we propose a convolutional neural network that predicts all pairwise distances between residues in an RNA, using a recently described smooth parametrization of Euclidean distance matrices. We achieve high-accuracy predictions on RNAs up to 100 nt in length in fractions of a second, a factor of 10<sup>7</sup> faster than existing molecular dynamics-based methods. We also convert our coarse-grained machine learning output into an all-atom model using discrete molecular dynamics with constraints. Our proposed computational pipeline predicts all-atom RNA models solely from the nucleotide sequence. However, this method suffers from the same limitation as nucleic acid molecular dynamics: the scarcity of available RNA crystal structures for training.</p>","PeriodicalId":8922,"journal":{"name":"Biophysical journal","volume":null,"pages":null},"PeriodicalIF":3.2000,"publicationDate":"2024-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11393712/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Biophysical journal","FirstCategoryId":"99","ListUrlMain":"https://doi.org/10.1016/j.bpj.2023.10.011","RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2023/10/14 0:00:00","PubModel":"Epub","JCR":"Q2","JCRName":"BIOPHYSICS","Score":null,"Total":0}
引用次数: 0
Abstract
Fast and accurate 3D RNA structure prediction remains a major challenge in structural biology, mostly due to the size and flexibility of RNA molecules, as well as the lack of diverse experimentally determined structures of RNA molecules. Unlike DNA structure, RNA structure is far less constrained by basepair hydrogen bonding, resulting in an explosion of potential stable states. Here, we propose a convolutional neural network that predicts all pairwise distances between residues in an RNA, using a recently described smooth parametrization of Euclidean distance matrices. We achieve high-accuracy predictions on RNAs up to 100 nt in length in fractions of a second, a factor of 107 faster than existing molecular dynamics-based methods. We also convert our coarse-grained machine learning output into an all-atom model using discrete molecular dynamics with constraints. Our proposed computational pipeline predicts all-atom RNA models solely from the nucleotide sequence. However, this method suffers from the same limitation as nucleic acid molecular dynamics: the scarcity of available RNA crystal structures for training.
期刊介绍:
BJ publishes original articles, letters, and perspectives on important problems in modern biophysics. The papers should be written so as to be of interest to a broad community of biophysicists. BJ welcomes experimental studies that employ quantitative physical approaches for the study of biological systems, including or spanning scales from molecule to whole organism. Experimental studies of a purely descriptive or phenomenological nature, with no theoretical or mechanistic underpinning, are not appropriate for publication in BJ. Theoretical studies should offer new insights into the understanding ofexperimental results or suggest new experimentally testable hypotheses. Articles reporting significant methodological or technological advances, which have potential to open new areas of biophysical investigation, are also suitable for publication in BJ. Papers describing improvements in accuracy or speed of existing methods or extra detail within methods described previously are not suitable for BJ.