Machine Learning in Preclinical Development of Antiviral Peptide Candidates: A Review of the Current Landscape

In the field of antiviral peptide (AVP) design, one of the most prominent limiting factors is the time and material cost required to perform the initial screening of novel AVPs. In particular, traditional target identification as well as traditional preclinical screening of novel drug candidates can be a very lengthy and expensive process. In recent decades, target identification and initial screening of AVPs has been increasingly carried out using machine learning (ML). The use of ML to initially screen potential interactions reduces the financial cost and lengthy time scale of preclinical AVP development, allowing for candidate peptides to be identified and screened faster, at a lower cost to both manufacturer and consumer. Additionally, the use of ML in generating and screening AVP candidates allows a more diverse chemical space to be explored than high-throughput screening methodologies allow. In silico generation and validation of AVP candidates also limits researcher contact with high BSL-rated viruses, thereby increasing the safety and accessibility of AVP design. This review seeks to provide a broad overview of the current uses of ML in early-stage AVP design, and to shed some light on the future direction of the field.