A SOFM approach to predicting HIV drug resistance

The self-organizing feature map (SOFM or SOM) neural network approach has been applied to a number of life sciences problems. In this paper, we apply SOFMs in predicting the resistance of the HIV virus to Saquinavir, an approved protease inhibitor. We show that a SOFM predicts resistance to Saquinavir with reasonable success based solely on the amino acid sequence of the HIV protease mutation. The best single network provided 69% coverage and 68% accuracy. We then combine a number of networks into various majority voting schemes. All of the combinations showed improved performance over the best single network, with an average of 85% coverage and 78% accuracy. Future research objectives are suggested based on these results.