TriNet

TriNet predicts anticancer peptides (ACPs) and antimicrobial peptides (AMPs) using a tri-fusion neural network that integrates serial fingerprints, sequence evolution, and physicochemical properties to improve peptide classification accuracy.

Key Features:

• Feature Integration: TriNet combines three distinct peptide feature sets—serial fingerprints, sequence evolution, and physicochemical properties—for comprehensive representation.
• Parallel Neural Network Modules: A convolutional neural network (CNN) module with channel attention, a bidirectional long short-term memory (BiLSTM) module, and an encoder module operate in parallel to extract spatial, sequential, and integrated representations.
• Iterative Training Approach: TriNet employs an iterative interaction strategy between samples in the training and validation datasets to optimize model learning.

Scientific Applications:

• Anticancer peptide prediction (ACPs): Predicts ACPs to support identification of peptide candidates for cancer therapeutic development.
• Antimicrobial peptide prediction (AMPs): Predicts AMPs to support identification of peptide candidates for treating infectious diseases.

Methodology:

Peptide data are processed through a tri-fusion architecture in which a CNN module with channel attention captures spatial patterns, a BiLSTM captures temporal sequence dependencies, an encoder integrates module outputs for classification, and an iterative interaction strategy is used between training and validation samples during training.