
DeepMRMP: A new predictor for multiple types of RNA modification sites using deep learning
Author(s) -
Ping Sun,
Yong Bing Chen,
Bo Liu,
Yan Gao,
Ye Han,
Fei He,
Jin Chao Ji
Publication year - 2019
Publication title -
mathematical biosciences and engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.451
H-Index - 45
eISSN - 1551-0018
pISSN - 1547-1063
DOI - 10.3934/mbe.2019310
Subject(s) - pseudouridine , rna , computational biology , feature (linguistics) , artificial intelligence , computer science , deep learning , transfer rna , machine learning , biology , gene , genetics , linguistics , philosophy
RNA modification plays an indispensable role in the regulation of organisms. RNA modification site prediction offers an insight into diverse cellular processing. Regarding different types of RNA modification site prediction, it is difficult to tell the most relevant feature combinations from a variant of RNA properties. Thereby, the performance of traditional machine learning based predictors relied on the skill of feature engineering. As a data-driven approach, deep learning can detect optimal feature patterns to represent input data. In this study, we developed a predictor for multiple types of RNA modifications method called DeepMRMP (Multiple Types RNA Modification Sites Predictor), which is based on the bidirectional Gated Recurrent Unit (BGRU) and transfer learning. DeepMRMP makes full use of multiple RNA site modification data and correlation among them to build predictor for different types of RNA modification sites. Through 10-fold cross-validation of the RNA sequences of H. sapiens, M. musculus and S. cerevisiae, DeepMRMP acted as a reliable computational tool for identifying N 1 -methyladenosine (m 1 A), pseudouridine (Ψ), 5-methylcytosine (m 5 C) modification sites.