Sequence-based Deep Learning Model for AMPs Activity Prediction Against Specific Pathogen Strains

Abstract

The pipeline of antimicrobial peptides (AMPs) discovery and design is costly and time-consuming. So, several machine learning approaches and techniques have been used in the in-silico stage to help discover the new AMPs by indicating their activity before moving them into the in-vitro and in-vivo stages. However, machine learning and statistical methods require a lot of feature engineering and domain experts. Recently, many deep learning approaches have shown their superior performance in several fields, which have inspired researchers to take advantage to reduce the efforts of AMPs discovery and design tasks. However, only some efforts provided a specific model that can determine the active AMPs against specific pathogens strains, which would improve the quality of samples that could pass the laboratory and clinical trials. This paper attempts to propose a sequence-base deep learning model that helps laboratories predict potential active AMPs against specific pathogen strains by preparing up-to-date datasets for activity AMPs based on the latest updated databases and taking the power of combining multiple deep learning architectures to build AMPs activity prediction models. The best AMPs activity prediction models achieved 97%-98% of accuracy and Matthew’s correlation coefficient of 0.93 to 0.96.