Prediction of coronavirus 3C-like protease cleavage sites using machine-learning algorithms

Area of Science:

• Virology
• Proteomics
• Bioinformatics

Background:

• The coronavirus 3C-like (3CL) protease is crucial for viral replication and immune evasion.
• Effective tools for identifying 3CL protease cleavage sites are currently limited.

Purpose of the Study:

• To investigate the diversity of 3CL protease cleavage sites across coronaviruses.
• To develop a computational model for predicting these cleavage sites.
• To identify potential human protein substrates of the 3CL protease.

Main Methods:

• Systematic analysis of coronavirus 3CL protease cleavage site motifs.
• Development and validation of a random forest (RF) model using amino acid indexes.
• Prediction of cleavage sites in viral and human proteins.
• Gene Ontology (GO) enrichment analysis of predicted human substrates.

Main Results:

• Highly conserved cleavage motifs were observed across Alphacoronavirus, Betacoronavirus, and Gammacoronavirus genera.
• The RF model achieved high accuracy (AUC 0.96 in cross-validation, 0.95 on independent test set).
• 1,352 human proteins were predicted as 3CL protease substrates, enriched in cytoskeleton-related functions.

Conclusions:

• A robust computational tool (3CLP webserver) for predicting 3CL protease cleavage sites has been developed.
• The findings offer insights into coronavirus molecular mechanisms and pathogenicity.
• Identified human protein substrates may represent novel targets for therapeutic intervention.