Predicting protein stability change upon single point mutation using multi-instance regression: a local conformational analysis approach

The prediction of stability change caused by a mutation in a protein structure is of vital importance for protein design and analysis. Several attempts have been made to predict these energy changes by analyzing the global conformational properties of a structure. To date, none of the research has focused solely on studying the effect of local conformational properties of a mutated residue to the final stability change. In my thesis I use multi-instance regression learning with output aggregation to learn and predict the energy change using the information from the local environment of the mutated residue. This research shows a high degree of correlation between the expected and predicted values of energy changes and a quantum leap from the current state-of-the-art.