sAMP-PFPDeep

sAMP-PFPDeep predicts short antimicrobial peptides (sAMPs; 30 or fewer amino acid residues) from peptide sequences to support discovery and validation of novel antimicrobial agents.

Key Features:

• Sequence encoding: Converts peptide sequences into image-like representations using three channels representing positional information, frequency data, and the summation of 12 key physicochemical features.
• Deep neural networks: Processes encoded images with image-based deep neural networks RESNET-50 and VGG-16 trained on a benchmark dataset derived from recent studies.
• Performance metrics: VGG-16 achieved 98.30% training accuracy and 87.37% testing accuracy; RESNET-50 achieved 96.14% training accuracy and 83.87% testing accuracy.
• Comparative analysis: Outperforms previously reported state-of-the-art methods in predicting sAMPs.
• Molecular docking validation: Employs molecular docking-based analysis to validate predictive results.

Scientific Applications:

• Antimicrobial peptide discovery: Identifies candidate short antimicrobial peptides for experimental validation and development.
• Computer-aided drug design: Provides predicted sAMP candidates for integration into peptide design and optimization workflows.
• Virtual screening protocols: Supports prioritization of peptides with therapeutic potential in virtual screening pipelines.

Methodology:

Peptide sequences are encoded into three-channel image-like representations (positional information, frequency data, and 12 physicochemical features), processed by RESNET-50 and VGG-16 models trained on a benchmark dataset, and validated using molecular docking-based analysis.