m5CStack: An integrated framework for m5C site prediction using multi-feature stacking.

Area of Science:

• * Molecular Biology
• * Genomics
• * Bioinformatics

Background:

• * RNA 5-methylcytosine (m5C) modifications are crucial for regulating RNA function.
• * High-throughput genomics generates vast data, challenging traditional m5C identification methods.
• * Computational tools are essential for efficient and accurate m5C site prediction.

Purpose of the Study:

• * To develop an advanced ensemble learning framework, m5CStack, for predicting RNA m5C modification sites.
• * To enhance the accuracy, robustness, and reliability of m5C site predictions.
• * To provide a user-friendly tool for RNA modification profiling.

Main Methods:

• * Developed m5CStack, an ensemble learning framework utilizing a stacking architecture.
• * Integrated multiple feature encoding techniques and machine learning models.
• * Evaluated performance on RNA datasets from *Homo sapiens*, *Mus musculus*, *Drosophila melanogaster*, and *Danio rerio*.
• * Employed SHAP-based analysis for feature significance.

Main Results:

• * m5CStack significantly outperformed existing prediction methods in accuracy, sensitivity, and specificity.
• * SHAP analysis identified key features contributing to prediction accuracy, enhancing model interpretability.
• * The framework demonstrated robust performance across diverse species.

Conclusions:

• * m5CStack is a powerful and accurate tool for RNA m5C modification site prediction.
• * The framework offers improved RNA modification profiling and insights into epigenetic regulation.
• * A web interface enhances accessibility for researchers worldwide.