Using triangulation to include target structure improves drug resistance prediction accuracy

Genomic data promises to provide individualized medicine. The cost of determining an individual genome will soon be low enough, if it isn´t already, to make genotyping a standard medical procedure. Unfortunately, using this data to predict phenotypic responses is a difficult and largely unsolved problem. Drug resistance in HIV is an excellent model system for this problem because there are many pairs of genotypes and associated drug-resistance phenotypes. There is also a great amount of structural and biochemical data on this system. We have found that including three-dimensional structural data for drug targets dramatically improves the quality of machine learning when applied to the association between genotype and phenotype.