A generative explainable model for antimicrobial peptide prediction using bidirectional temporal convolutional neural network

The computational identification of antimicrobial peptides (AMPs) is particularly challenging, given the multi-faceted inherent structural and functional characteristics that govern their activity. Current models for predicting properties of AMPs fail to sufficiently represent the relationship between the characteristics of AMPs, and hence continue to restrict the innovative discovery of peptides for use in targeted cancer therapy. We present GAC-BiTCNN-AMP, the first of its kind hybrid generative and explainable deep learning framework, to address these challenges. GAC-BiTCNN-AMP combines a Generative Adversarial Network to improve data diversity, Capsule Networks to explain what the model has learned about both linear and non-linear hierarchical relationships in the data, and a Bidirectional Temporal Convolutional Neural

Network to learn the relationships among and between the elements of the data in a given context. We also use embedding representations of three state-of-the-art protein language models, ProtTrans-T5, UniRep, and ESM-2, combined with a novel evolutionary descriptor, PsePSSM-DCT, to improve the representation of a biological signal of the data. The model also uses a wrapper with XGBoost for Forward Feature Selection to limit the number of model features to the most relevant sequence features. These various model features provide the best balance between predictability and explainability for AMPs and demonstrates great predictive potency.

Message: Hi FE, I need help with “A generative explainable model for antimicrobial peptide prediction using bidirectional temporal convolutional neural network” in “Biotechnology & Biomedical Engineering”