Datt-AVP: Antiviral Peptide Prediction by Sequence-Based Dual Channel Network With Attention Mechanism

Abstract

With the frequent outbreak of viral pandemics, the search for efficient antiviral drugs has become an urgent task. Antiviral peptides (AVPs) have been proven to prevent the infection of host cells by viruses. This study proposes a novel tool called Datt-AVP for AVP prediction based on the peptide sequences. A dual-channel deep learning model of Long Short-Term Memory network and convolutional neural network, along with a pre-trained protein language model as a feature extractor, was applied to capture features in the antiviral peptide sequences simultaneously. Furthermore, self-attention module was added to promote the performance of our prediction results. It showed good recognition ability for antiviral peptides with different lengths. Our tool achieved an accuracy rate of 96.1% on the benchmark dataset, which out-performed state-of-the-art tools.