MSCMamba: Prediction of Antimicrobial Peptide Activity Values by Fusing Multiscale Convolution with Mamba Module

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Antimicrobial peptides (AMPs) show promise as novel antibiotics.
• Existing methods primarily focus on qualitative prediction of AMP activity, with limited quantitative prediction capabilities.

Purpose of the Study:

• To develop a quantitative prediction model for AMP activity values.
• To introduce MSCMamba, a regression model integrating multiscale convolutional neural networks and Mamba modules.

Main Methods:

• Feature extraction from AMP sequences using multiple encoding techniques.
• Utilizing a multiscale convolutional network for local feature capture and a Mamba module for long-range dependency analysis.
• Fusing extracted features and predicting AMP activity values via a linear layer.

Main Results:

• MSCMamba demonstrated superior performance compared to existing state-of-the-art methods.
• Achieved a 10.66% improvement in R-squared value, increasing it from 0.422 to 0.467.
• Ablation experiments confirmed the model's components' validity and the benefits of feature diversification.

Conclusions:

• MSCMamba offers a novel and accurate approach for quantitative prediction of AMP activity.
• The model is expected to accelerate the discovery and development of new antibiotic agents.