Residue Adjacency Matrix Based Feature Engineering for Predicting Cysteine Reactivity in Proteins

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Reactive oxygen and nitrogen species cause oxidative damage to cellular components.
• This damage is implicated in chronic diseases such as diabetes, cancer, Parkinson's, and heart disease.
• Cysteine residues in proteins are particularly susceptible to oxidation.

Purpose of the Study:

• To develop a fast and inexpensive computational method for identifying oxidizable cysteine residues in proteins.
• To improve upon existing methods for predicting cysteine oxidation.
• To aid in understanding the role of cysteine oxidation in disease pathogenesis.

Main Methods:

• Utilized machine learning algorithms for classification.
• Developed novel features: RAMmod and RAMseq.
• Incorporated established features: PROPKA, SASA, PSS, and PSSM.
• Employed template matching with MODELLER for 3D coordinate acquisition for feature extraction.

Main Results:

• The developed algorithm, RAM, demonstrated a significant mean improvement of 22.04% in Matthews Correlation Coefficient (MCC) over N6C (p=0.015).
• RAM showed a significant increase in MCC over PSSM (p=0.040), with an average improvement of 70.09%.
• The algorithm requires only the protein sequence as input.

Conclusions:

• Machine learning offers a rapid and cost-effective alternative to traditional wet lab methods for identifying oxidizable cysteines.
• The RAM algorithm shows significant potential for advancing research in oxidative stress and related chronic diseases.
• Accurate identification of oxidized cysteines can enhance our understanding of disease mechanisms.