Modeling multi-species RNA modification through multi-task curriculum learning.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• N6-methyladenosine (m6A) is the most abundant mRNA modification in eukaryotes.
• m6A regulates crucial biological processes including gene expression, RNA stability, and translation.
• Existing computational methods for m6A prediction lack accuracy, interpretability, and cross-species generalizability.

Purpose of the Study:

• To develop a novel, interpretable computational framework (MASS) for predicting m6A modification sites.
• To capture m6A features across multiple species simultaneously using a multi-task curriculum learning strategy.
• To improve the accuracy, interpretability, and generalizability of m6A prediction models.

Main Methods:

• Implementation of a multi-task curriculum learning strategy within the MASS framework.
• Training and validation of the MASS model on cross-species m6A datasets.
• Analysis of contextual sequence features identified by MASS for m6A site prediction.

Main Results:

• MASS demonstrated superior performance compared to existing state-of-the-art m6A prediction methods.
• The framework successfully captured interpretable sequence features, correlating with known RNA-binding protein motifs.
• MASS facilitated the elucidation of similarities and differences in m6A features across species.

Conclusions:

• MASS provides a powerful and interpretable tool for characterizing m6A modification landscapes.
• The framework aids in understanding the regulatory code of m6A and its impact on gene regulation.
• MASS enables deeper insights into m6A's role in gene expression, RNA stability, translation, and RNA structure.