Computational prediction of phosphorylation sites of SARS-CoV-2 infection using feature fusion and optimization...

Area of Science:

• Virology
• Computational Biology
• Biochemistry

Background:

• SARS-CoV-2 has caused a global health and economic crisis.
• Understanding viral phosphorylation sites is crucial for deciphering infection mechanisms and host cell alterations.
• Existing computational tools for phosphorylation site prediction are limited, often focusing only on specific residues (S/T or Y).

Purpose of the Study:

• To develop a novel computational predictor for identifying all residue (serine, threonine, and tyrosine) phosphorylation sites in SARS-CoV-2.
• To enhance the accuracy and scope of phosphorylation site prediction beyond currently available methods.

Main Methods:

• Extraction of ten distinct feature descriptors based on composition, evolutionary, and position-specific information.
• Evaluation of feature descriptor discriminative power using five different classifiers.
• Development of the final prediction model, LGB-IPs, using Light Gradient Boosting (LGB) and the top two integrated features.

Main Results:

• Light Gradient Boosting (LGB) demonstrated superior performance among the tested classifiers.
• Five feature descriptors exhibited excellent discriminative capabilities.
• The developed LGB-IPs model achieved high performance metrics: 0.831 ACC, 0.662 MCC, and 0.907 AUC on 10-fold cross-validation.
• The model's performance was validated through independent evaluation, showing consistent results.

Conclusions:

• The novel LGB-IPs predictor effectively identifies all residue phosphorylation sites in SARS-CoV-2.
• The approach provides valuable insights into SARS-CoV-2 phosphorylation mechanisms relevant to biomedical researchers.
• This tool can aid in understanding viral infection and host cellular processes during SARS-CoV-2 infection.