Global analysis of N6-methyladenosine functions and its disease association using deep learning and network-based methods

Abstract

N6-methyladenosine (m 6 A) is the most abundant methylation, existing in >25% of human mRNAs. Exciting recent discoveries indicate the close involvement of m 6 A in regulating many different aspects of mRNA metabolism and diseases like cancer. However, our current knowledge about how m 6 A levels are controlled and whether and how regulation of m 6 A levels of a specific gene can play a role in cancer and other diseases is mostly elusive. We propose in this paper a computational scheme for predicting m 6 A-regulated genes and m 6 A-associated disease, which includes Deep-m 6 A, the first model for detecting condition-specific m 6 A sites from MeRIP-Seq data with a single base resolution using deep learning and a new network-based pipeline that prioritizes functional significant m 6 A genes and its associated diseases using the Protein-Protein Interaction (PPI) and gene-disease heterogeneous networks. We applied Deep-m 6 A and this pipeline to 75 MeRIP-seq human samples, which produced a compact set of 709 functionally significant m 6 A-regulated genes and nine functionally enriched subnetworks. The functional enrichment analysis of these genes and networks reveal that m 6 A targets key genes of many critical biological processes including transcription, cell organization and transport, and cell proliferation and cancer-related pathways such as Wnt pathway.The m 6 A-associated disease analysis prioritized five significantly associated diseases including leukemia and renal cell carcinoma. These results demonstrate the power of our proposed computational scheme and provide new leads for understanding m 6 A regulatory functions and its roles in diseases.Keywords: N6-methyladenosine (m 6 A), methylated RNA immunoprecipitation sequencing (MeRIP-Seq ), convolutional neural networks, m 6 A site prediction, m 6 A functional prediction, m 6 A-disease association Author summaryThe goal of this work is to identify functional significant m 6 A-regulated genes and m 6 A-associated diseases from analyzing an extensive collection of MeRIP-seq data. To achieve this, we first developed Deep-m 6 A, a CNN model for single-base m 6 A prediction. To our knowledge, this is the first condition-specific single-base m 6 A site prediction model that combines mRNA sequence feature and MeRIP-Seq data. The 10-fold cross-validation and test on an independent dataset show that Deep-m 6 A outperformed two sequence-based models. We applied Deep-m 6 A followed by network-based analysis using HotNet2 and RWRH to 75 human MeRIP-Seq samples from various cells and tissue under different conditions to globally detect m 6 A-regulated genes and further predict m 6 A mediated functions and associated diseases. This is also to our knowledge the first attempt to predict m 6 A functions and associated diseases using only computational methods in a global manner on a large number of human MeRIP-Seq samples. The predicted functions and diseases show considerable consistent with those reported in the literature, which demonstrated the power of our proposed pipe...

m6A demethylase ALKBH5 inhibits pancreatic cancer tumorigenesis by decreasing WIF-1 RNA methylation and mediating Wnt signaling

“…Accumulating evidence has showed the epigenetic regulation of Wnt/β-catenin signaling in cancer, though, the impact of mRNA modification is still insufficiently studied. Until very recently, global analysis of m 6 A functions in 75 MeRIP-seq human samples using deep learning and network-based methods identified 709 functionally significant m 6 A-regulated genes, which were enriched in many critical biological processes including cancer-related pathways such as Wnt pathway [30]. YTH N 6 -methyladenosine RNA binding protein 1 (YTHDF1) regulates tumorigenicity and cancer stem cell-like activity in human colorectal carcinoma by mediating Wnt/β-catenin pathway [31].…”

m6A Reader: Epitranscriptome Target Prediction and Functional Characterization of N6-Methyladenosine (m6A) Readers

“…Furthermore, potential RNA methylation-disease associations have been revealed by m 6 Avar ( Zheng et al, 2018 ) and m 6 ASNP ( Jiang et al, 2018 ). With a similar purpose, heterogeneous networks have been used in DRUM ( Tang et al, 2019 ), FunDMDeep-m 6 A ( Zhang et al, 2019b ) and Deepm 6 A ( Zhang et al, 2019a ), showing a new perspective in studying disease-associated RNA methylation.…”

Bioinformatics approaches for deciphering the epitranscriptome: Recent progress and emerging topics

“…The most widely used deep learning models are convolutional neural network (CNN), which can effectively learn the motif related features from RNA sequence, and recurrent neural network (RNN), which can learn the non-linear sequential features from RNA sequence, including long short-term memory unit (LSTM) and gated recurrent unit (GRU). Gene2vec [124] , DeepPromise [122] , iN6-Methyl (5-step) [125] and Deep-m6A [126] built CNN models to predict m 6 A or m 1 A modifications; BERMP [127] employed a bidirectional Gated Recurrent Unit (BGRU) model to predict m 6 A; DeepMRMP [128] adopted bidirectional Gated Recurrent Unit (BGRU) and transfer learning to predict m 6 A, m 1 A, pseudouridine and m 5 C; the DL models of DeepM6ASeq [129] consists of two layers of CNN, one bidirectional long short-term memory (BLSTM) layer and one fully connected (FC) layer.…”