Determining Positions Associated with Drug Resistance on HIV-1 Proteins: A Computational Approach

The computational modeling of HIV-1proteins has become a useful framework allowing understanding the virus behavior (e.g. mutational patterns, replication process or resistance mechanism). For instance, predicting the drug resistance from genotype means to solve a complicated sequence classification problem. In such kind of problems proper feature selection could be essential to increase the classifiers performance. Several sequence positions that have been previously associated with resistance are known, although we believe that other positions could be discovered. More explicitly, we observed that using positions reported in the literature for the reverse transcriptase protein, the final decision system exhibited inconsistent mutations. However, finding a minimal subset of features characterizing the whole sequence involve a challenging combinatorial problem. This research proposes a model based on Variable Mesh Optimization and Rough Sets Theory for computing those sequence positions associated with resistance, leading to more consistent decision systems. Finally, our model is validated across eleven well-known reverse transcriptase inhibitors.