Application of a Combined Approach for Predicting a Peptide-Protein Binding Affinity Using Regulatory Regression Methods with Advance Reduction of Features

The paper proposes a phased method of applying filtering algorithms, descriptor clustering. At the first stage, the features are reduced by sequential application of the moving average and FFT filtering algorithms and the reduction of the discretization step. At the second stage, for the selection of signs using the cluster analysis method X-means. At the final stage, regression models are constructed using the regulatory regression algorithms L1, L2, and Leastsquares. The resulting models are highly accurate, robust and adequate. In general, the work proposed a new method for predicting the binding affinity of peptides in order to find the numerical values of peptide bonds.