m6ACali: machine learning-powered calibration for accurate m6A detection in MeRIP-Seq

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• N6-methyladenosine (m6A) is a crucial epitranscriptomic modification.
• MeRIP-Seq is widely used for m6A profiling but can be affected by non-specific antibody enrichment.
• Accurate m6A site identification is essential for understanding its biological roles.

Purpose of the Study:

• To develop a machine learning framework, m6ACali, to improve the accuracy of m6A epitranscriptome profiling.
• To reduce the impact of non-specific antibody binding in MeRIP-Seq data.
• To provide a universal method for enhancing m6A profiles.

Main Methods:

• Development of m6ACali, a genomic feature-based classifier.
• Utilizing machine learning to distinguish true m6A sites from those detected in in-vitro transcribed (IVT) controls.
• Testing m6ACali on novel MeRIP-Seq datasets without paired IVT controls.

Main Results:

• m6ACali effectively identifies non-specific binding peaks identified by exomePeak2 and MACS2.
• Off-target binding sites are associated with short exons, short mRNAs, and high read coverage regions with similar motifs.
• The ML strategy can adjust differentially methylated peaks and other antibody-dependent m6A detection techniques.

Conclusions:

• m6ACali significantly enhances the accuracy of m6A profiling in MeRIP-Seq experiments.
• The framework offers a robust method for identifying true m6A sites, even without IVT controls.
• m6ACali sets a new benchmark for omics-level m6A data integration and analysis.