Prediction of folding type of proteins using mixed-integer linear programming

Abstract

Proteins are classified into four main structural classes by considering their amino acid compositions. Traditional approaches that use hyperplanes to partition data sets into two groups perform poorly due to the existence of four classes. Therefore, a novel method that uses mixed-integer programming is developed to overcome difficulties and inconsistencies of these traditional approaches. Mixed-integer programming (MIP) allows the use of hyper-boxes in order to define the boundaries of the sets that include all or some of the points in that class. For this reason, the efficiency and accuracy of data classification with MIP approach can be improved dramatically compared to the traditional methods. The efficiency of the proposed approach is illustrated on a training set of 120 proteins (30 from each type). The prediction results and their validation are also examined.