Multi-feature fusion for gene prediction and functional peptide identification

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Anticancer peptides (ACPs) and antimicrobial peptides (AMPs) show therapeutic promise but require efficient identification methods.
• Wet-lab identification is slow and costly, hindering high-throughput drug screening.
• Existing computational tools lack robust feature representation and cross-task prediction capabilities.

Purpose of the Study:

• To develop an advanced computational tool, GP2FI, for accurate functional peptide prediction.
• To improve the identification of anticancer peptides and antimicrobial peptides for therapeutic applications.
• To address the limitations of current methods in feature extraction and cross-task prediction.

Main Methods:

• GP2FI employs a two-stage deep learning architecture: MHA-preconv for gene prediction and FuncPred-CB for functional peptide identification.
• MHA-preconv combines Convolutional Neural Networks (CNNs) with Transformer encoder layers to capture sequence patterns and dependencies.
• FuncPred-CB utilizes a pre-trained BERT language model to extract contextual semantic features from amino acid sequences.

Main Results:

• GP2FI demonstrates superior performance over state-of-the-art methods on benchmark datasets.
• The MHA-preconv model effectively identifies coding regions, improving subsequent functional prediction.
• GP2FI achieves high accuracy and robust performance metrics in peptide identification tasks.

Conclusions:

• GP2FI offers a significant advancement in computational peptide identification for drug discovery.
• The tool's deep learning architecture enhances feature representation and prediction accuracy.
• GP2FI provides an efficient and effective alternative to traditional wet-lab methods for identifying therapeutic peptides.