m

Objective The most prevalent mRNA modification, N6-methyladenosine (m6A) plays an important role in various RNA metabolism, including gene expression and translation. By recruiting different “reader” proteins and their cofactors, m6A modification can affect messenger RNA (mRNA) degradation, splicing, nuclear export and translation. However, the selective mechanism by which m6A sites regulate mRNA translation through m6A reader YTHDF1 binding remains poorly understood, due to a lack of computational methods for identifying context-specific m6A sites that regulate translation. To address this, we developed a novel computational framework named m6ATEpre, the first tool designed to predict cell-specific m6A sites that regulate translation efficiency. Methods m6ATEpre integrates multi-omics data, introduces a novel feature representation strategy for m6A site sequences, and employs an autoencoder to effectively capture embedded feature representations. Specifically, m6ATEpre first integrated MeRIP-seq data and PAR-CLIP data through overlapping m6A sites with YTHDF1 binding sites and identified YTHDF1-mediated m6A sites. Then, m6ATEpre detected the translation gene by analyzing the Ribo-seq data under YTHDF1 knockdown vs control condition. Genes whose translation is mediated by YTHDF1 in an m6A-dependent manner were identified by a significant decrease in translation efficiency upon YTHDF1 knockdown. Next, we proposed a binary vector indicating the presence or absence of YTHDF1 binding motifs to characterize each m6A site sequence. This represents a novel feature representation strategy for m6A sites. m6ATEpre utilized the autoencoder to extract the potentially important feature representations and constructed a multilayer perceptron neural networks model to predict potential m6A sites that regulating translation efficiency. Results A comprehensive evaluation of m6ATEpre was conducted through a series of experiments. We compared its performance against that of a similar prediction task model, as well as other classifiers. The results indicate that m6ATEpre achieved the best prediction performance. In addition, we analyzed different feature representation strategies and performed ablation experiments to validate the rationality of the model design. The results demonstrate that our proposed feature representation strategy has a greater advantage in improving prediction performance. In the HeLa cell line, bioinformatic analysis of the metagene distribution and sequence minimum free energy of m6A sites regulating translation efficiency (m6A-reg-TE sites) revealed their specific properties in translation regulation. Functional enrichment analysis indicated that m6A-reg-TE genes are associated with specific biological processes and KEGG pathways. By integrating the binding sites of YTHDF1 co-factors with m6A-reg-TE sites, we revealed that YTHDF1-mediated and m6A-dependent translation efficiency regulation requires the cooperation of multiple translation-regulatory RNA-binding proteins among its co-factors in the HeLa cell line. Furthermore, we extended our predictions to the dataset of the HEK293T cell line. Similarly, bioinformatic analysis of the metagene distribution and functional enrichment revealed the cell-specific characteristic of these predicted m6A-reg-TE sites in HEK293T cells. Likewise, integrated analysis of multiple YTHDF1 co-factors and m6A-reg-TE sites predicted in the HEK293T cell line reveals their m6A-dependent cooperation in regulating translation efficiency. Conclusion m6ATEpre is a timely tool that will advance our understanding of the mechanisms of m6A regulation in translation efficiency. The source code and datasets used in this work can be downloaded from https://www.scidb.cn/s/bAZZFr.