TransACVP: A Transformer-Based Predictive Model for Identifying Anti-Coronavirus Peptides

Since the outbreak of the Corona Virus Disease 2019 (COVID-19) in 2019, human beings have been struggling with this epidemic. Currently, the novel coronavirus still poses a threat to human health. Therefore, developing effective drugs to suppress or treat related diseases caused by the coronavirus remains an urgent need. Studies have shown that many peptide inhibitors exhibit effective inhibitory activity against the coronavirus. In this study, we collected anti-coronavirus peptide sequences and constructed 24 prediction models using four peptide descriptors and six feature selection methods, combined with four traditional machine learning methods. We also constructed three predictive models using three deep learning methods. By comparing the performance of the 27 prediction models in a five-fold cross-validation, we ultimately developed a Transformer-based predictor for identifying anti-coronavirus peptides, named TransACVP. In a five-fold cross-validation, the accuracy, sensitivity, specificity, Matthew's correlation coefficient, and area under the receiver operating characteristic curve of TransACVP are 0.8879, 0.8667, 0.9091, 0.7791, and 0.9163, respectively.