PepMCP: a graph-based membrane contact probability predictor for membrane-lytic antimicrobial peptides

Area of Science:

• Computational biology
• Biophysics
• Drug discovery

Background:

• The membrane-lytic mechanism of antimicrobial peptides (AMPs) is crucial but often overlooked in computational discovery due to limitations in predicting membrane-binding propensity.
• Previous methods for predicting membrane contact probability (MCP) were not optimized for short peptides targeting bacterial membranes, limiting their effectiveness.

Purpose of the Study:

• To develop a tailored model, PepMCP, for accurately predicting the membrane contact probability (MCP) of short antimicrobial peptides (AMPs).
• To enhance the in silico discovery of membrane-lytic AMPs by providing a reliable metric for their membrane-binding propensity.

Main Methods:

• Collected over 500 membrane-lytic AMPs from scientific literature.
• Performed coarse-grained molecular dynamics (MD) simulations to determine residue-level MCP values.
• Trained the PepMCP model using the GraphSAGE framework, representing peptide sequences as graphs.

Main Results:

• PepMCP achieved high accuracy with a Pearson correlation coefficient of 0.883 and an RMSE of 0.123 on the test set.
• The model effectively identifies membrane-lytic AMPs based on predicted MCP values.
• A comprehensive database, MemAMPdb, and a user-friendly web server were established.

Conclusions:

• PepMCP offers a significant advancement in predicting the membrane-binding propensity of short antimicrobial peptides.
• This tool facilitates mechanism-driven discovery of novel membrane-lytic AMPs.
• The availability of PepMCP and MemAMPdb supports further research in antimicrobial peptide development.