Computational Method for Identifying Malonylation Sites by Using Random Forest Algorithm.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics

Background:

• Protein malonylation is a recently discovered post-translational modification impacting metabolic pathways and diseases.
• Existing computational methods for predicting malonylation sites struggle with imbalanced datasets.

Purpose of the Study:

• To develop a novel computational method for predicting protein malonylation sites.
• To address the challenge of imbalanced data in malonylation site prediction using the synthetic minority over-sampling technique (SMOTE).

Main Methods:

• Utilized four feature types: amino acid composition, Position-Specific Scoring Matrix (PSSM), amino acid factor, and disorder.
• Employed a two-step feature selection process (Max-Relevance Min-Redundancy and Incremental Feature Selection) with a Random Forest algorithm.
• Developed an optimal classifier based on the selected feature subset.

Main Results:

• Achieved an F1-measure of 0.356 with the optimal classifier.
• Identified specific PSSM and disorder features as potentially significant indicators of malonylation sites.

Conclusions:

• The developed computational method offers a promising approach for predicting malonyllysine.
• The findings provide insights into the molecular mechanisms underlying protein malonylation.