A Review of Machine Learning Approaches for Predicting Viral Evolution

Abstract

Viral evolution is a complex and natural process that determine how viruses transmit, evolve, and cause disease. Understanding how viruses evolve and mutate is important from a public health perspective to guide vaccine development, early diagnosis and outbreak response. Traditional approaches of studying viral evolution (i.e., phylogenetic analysis, epidemiologic modeling) can struggle to keep pace with the rapid and often unpredictable nature of viral evolution. Recently, machine learning (ML) has emerged as a promising means of analyzing the vast amount of data created through viral genomics that can help distinguish evolutionary patterns and predict mutations. This paper reviews some of the machine learning techniques being utilized in the study of viral evolution through applied machine learning, with a focus on machine learning model components and procedures (i.e., no virological basis). We review model types used in the analysis of viral prediction, including deep learning architectures, such as, convolutional neural networks (CNNs), long short term memory (LSTM) networks, transformers and traditional modeling approaches in machine learning such as support vector machines (SVMs) and hybrid approaches. In addition to models study, we also review data sources (e.g GISAID, GenBank, NCBI) and how data was implemented to feed the model. Next we will present evaluation metrics used to validate and evaluate models. Finally, we discuss future issues and directions for machine learning and viral evolution prediction methods including data size, model generalizability, computational issues and model interpretability. Overall, these advancements with respect to ML and viral evolution research have made important inroads to the modeling of viral evolution prediction. However, issues such as stability, model types, interpretability, and real world applications remain.