Docking features for predicting binding loss due to protein mutation

The human genome contains a large number of protein polymorphisms due to individual genome variation. How many of these polymorphisms lead to altered protein-protein interaction is unknown. We have developed a method that uses docking simulations to predict whether variants have altered interactions with their binding partners. A novel docking score normalization that compares the docking of mutant-containing protein pairs to that of the wild-type pair is introduced. Using the SKEMPI database and CAPRI, a training set of 167 mutant pairs (87 binders, 80 non-binders) were identified and docked using the docking program, HADDOCK. A random forest classifier that uses the differences in resulting docking scores for the 167 mutant pairs, to distinguish between variants that have either completely or partially lost binding ability, was used. 50% of non-binders were correctly predicted with a false discovery rate of only 2%. This allows for the rapid identification of a large number of protein polymorphisms that are likely to have a physiological consequence. The model was tested on a set of 15 HIV-1 - human, as well as 7 human - human glioblastoma-related, mutant proteins pairs: 50% of combined non-binders were correctly predicted with a false discovery rate of 10%.