Computational modeling and prediction of the human immunodeficiency virus (HIV) strains

Abstract

This paper describes a stochastic approach for modeling the changes observed in the DNA sequence of a highly mutating virus, such as the human immunodeficiency virus (HIV). This modeling process is begun by clustering the known DNA sequences from the virus population into groups such that the individual clusters represent biological strains of the modeled virus. Next, a hidden Markov model (HMM) is associated with each cluster, and its parameters computed using Baum-Welch's expectation maximization procedure. In this manner, the sequences within a cluster represent a maximally likely random sample drawn from the learned HMM. After the HMM for each strain cluster has thus been learned, it can further be used to generate additional samples of viral DNA sequences that are expected from the same underlying HMM. These newly predicted sequences would represent a maximally likely set of sequences belonging to a given viral strain modeled by the underlying HMM.