Dynamical interaction analysis of proteins by a random forest-fragment molecular orbital (RF-FMO) method and application to Src tyrosine kinase

Identifying key intermolecular (amino acid) interactions is crucial for understanding intrinsic protein functions. In this study, we established an efficient method for discovering key interactions by combining the random forest (RF) method, a machine learning algorithm, and an interaction analysis based on the fragment molecular orbital (FMO) method.We applied this method to Src tyrosine kinase and verified its efficacy. We performed molecular dynamics simulations of both the open and closed forms of Src and selected 50 snap-shots for each. Then, pair interaction energy (PIE) or inter -fragment interaction energy (IFIE) analyses were performed using FMO with the van der Waals (vdW)-corrected density functional tight-binding (DFTB) method. Among the 100 (c) 34453 data sets, we can identify the key amino acid pair regu-lating the open-close transition. This is consistent with the experimental and theoretical results, indicating the usefulness of the presented method. In contrast to the conventional FMO PIE interaction analysis, in the proposed method, the protein dynamics can be partially included using hundreds of trajectory data.