m6Aminer: Predicting the m6Am Sites on mRNA by Fusing Multiple Sequence-Derived Features into a CatBoost-Based...

Area of Science:

• Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• N6-methyladenosine (m6A) is a crucial post-transcriptional modification impacting mRNA stability and cancer progression.
• Accurate identification of m6A sites is vital for understanding its biological roles and medical applications.
• Current experimental methods for m6A site identification are costly and time-consuming, hindering large-scale analysis.

Purpose of the Study:

• To develop an efficient computational tool, m6Aminer, for identifying m6A sites on mRNA.
• To leverage machine learning, specifically CatBoost, for high-throughput m6A site prediction.
• To establish a foundation for comprehensive functional research of m6A modifications.

Main Methods:

• Employed nine distinct feature encoding schemes to construct an initial feature space for m6A site prediction.
• Utilized the ExtraTreesClassifier algorithm for feature importance ranking, selecting the top 300 features.
• Developed a CatBoost-based prediction model, m6Aminer, for identifying m6A sites.

Main Results:

• m6Aminer achieved an average AUC of 0.913 (10-fold cross-validation) and 0.754 (independent test).
• Demonstrated competitive performance compared to existing state-of-the-art models like m6AmPred and DLm6Am.
• The selected optimal feature subset significantly contributed to the model's predictive accuracy.

Conclusions:

• m6Aminer provides a robust and efficient computational approach for identifying m6A sites.
• The tool facilitates large-scale transcriptome-wide m6A site identification, accelerating research.
• This work lays the groundwork for deeper functional investigations into the role of m6A in biological processes and diseases.