A machine learning approach for predicting methionine oxidation sites.

Area of Science:

• Biochemistry
• Computational Biology
• Proteomics

Background:

• Methionine oxidation to methionine sulfoxide was traditionally viewed as damage.
• Emerging evidence suggests this reversible reaction acts as a regulatory post-translational modification.
• Methionine sulfoxidation may play a role in cellular redox regulation, prompting proteomic investigations.

Purpose of the Study:

• To develop computational models for predicting methionine oxidation sites.
• To offer an alternative to expensive and time-consuming experimental proteomic studies.
• To investigate the structural context influencing methionine oxidation.

Main Methods:

• Developed predictive models using random forests, support vector machines, and neural networks.
• Created a curated dataset of 113 polypeptides with 975 methionyl residues.
• Utilized machine learning on features including solvent accessible area and residue proximity.

Main Results:

• Identified key predictive features: solvent accessible area, distance to N-terminal methionine, and proximity to aromatic residues.
• Random forests model achieved the highest performance.
• Achieved accuracy of 0.7468±0.0567, sensitivity of 0.6817±0.0982, and specificity of 0.7557±0.0721.

Conclusions:

• Presented the first predictive models for in vivo methionine oxidation sites.
• Models offer insights into structural determinants of methionine oxidation-proneness.
• Models can prioritize methionyl residues for studying regulatory post-translational modifications.