A deep learning framework to predict binding preference of RNA constituents on protein surface

• Published in: Nature communications 2019-10, Vol.10 (1), p.4941-13, Article 4941
• Main Authors: Lam, Jordy Homing, Li, Yu, Zhu, Lizhe, Umarov, Ramzan, Jiang, Hanlun, Héliou, Amélie, Sheong, Fu Kit, Liu, Tianyun, Long, Yongkang, Li, Yunfei, Fang, Liang, Altman, Russ B., Chen, Wei, Huang, Xuhui, Gao, Xin
• Format: Article
• Language: English
• Subjects: 631/114/1305; 631/45/500; Adenine - metabolism; Animals; Area Under Curve; Argonaute 2 protein; Argonaute Proteins - metabolism; Binding; Constituents; Cytosine - metabolism; Deep Learning; Gene Knockdown Techniques; Gene regulation; Gene sequencing; Guanine - metabolism; Humanities and Social Sciences; Humans; Immunoprecipitation; Machine learning; Mice; multidisciplinary; Phosphates - metabolism; Post-transcription; Protein Binding; Protein structure; Proteins; Ribonuclease III; Ribonuclease III - metabolism; Ribose - metabolism; RNA - metabolism; RNA, Small Interfering; RNA-binding protein; RNA-Binding Proteins - metabolism; ROC Curve; Science; Science (multidisciplinary); siRNA; Uracil - metabolism
• ISSN: 2041-1723; 2041-1723
• DOI: 10.1038/s41467-019-12920-0

Protein-RNA interaction plays important roles in post-transcriptional regulation. However, the task of predicting these interactions given a protein structure is difficult. Here we show that, by leveraging a deep learning model NucleicNet, attributes such as binding preference of RNA backbone constituents and different bases can be predicted from local physicochemical characteristics of protein structure surface. On a diverse set of challenging RNA-binding proteins, including Fem-3-binding-factor 2, Argonaute 2 and Ribonuclease III, NucleicNet can accurately recover interaction modes discovered by structural biology experiments. Furthermore, we show that, without seeing any in vitro or in vivo assay data, NucleicNet can still achieve consistency with experiments, including RNAcompete, Immunoprecipitation Assay, and siRNA Knockdown Benchmark. NucleicNet can thus serve to provide quantitative fitness of RNA sequences for given binding pockets or to predict potential binding pockets and binding RNAs for previously unknown RNA binding proteins. Interactions between proteins and RNA are an important mechanism for post-transcriptional regulation, but predicting these interactions is difficult. Through a deep learning approach, here the authors predict RNA-binding sites and binding preference based on the local physicochemical properties of the protein surface.