Using a temporal Bayesian approach in order to nd mutational networks in HIV
Pablo Hernández-Leal, Alma Rios-Flores, Santiago Ávila‐Ríos, Gustavo Reyes-Ter, Jesus A. Gonz, Felipe Orihuela‐Espina, Eduardo F. Morales, Luis Enrique Sucar · 2011
Much of the HIV (Human Immunodeciency Virus) success is due to its fast evolution which makes it drug ressistant. Understanding viral evolution and its relation to pharmacology is of utmost importance in ghting diseases caused by the HIV. Although the mutations conferring drug resistance are mostly known, the dynamics of the appearance chain of those mutations remains poorly understood. Here, we apply a Temporal Nodes Bayesian Network (TNBN) to data extracted from the HIV Stanford database to explore the probabilistic relationships between mutations and antiretrovirals. We aim to unveil existing mutation networks and establish their probabilistic formation sequence. The model predictive accuracy is evaluated in terms of the relative Brier score, relative temporal error, and total number of intervals. We developed models for the protease, an important protein of the HIV. The learned models captured well known relationships, which qualitatively provide a validity measure. In our experiments, two previously unknwon mutational networks were retrieved and their probabilistic temporal sequentiation uncovered. With our results, we demonstrate the usefulness of TNBN for studying drug-mutation and mutation-mutation networks. We