In Silico Identification of TYK2 Pseudokinase Inhibitors Using Machine Learning and MD Simulations

Abstract The development of small‐molecule inhibitors for Janus kinases (JAKs) has transformed the treatment of autoimmune disorders. However, developing isoform‐selective inhibitors that specifically target the catalytically active regions of this kinase family is still a substantial issue. Recently, selectively inhibiting tyrosine kinase 2 (TYK2) has emerged as a promising strategy for treating a variety of autoimmune diseases. Deucravacitinib, an innovative oral medication, targets a critical molecule implicated in the pathogenesis of psoriasis by selectively inhibiting the TYK2 pseudokinase domain. Therefore, in this study, a medication repurposing method was used to identify more selective TYK2 inhibitors by machine learning (ML), structure based virtual screening, and molecular dynamics simulations. Initially, a random forest regression model from the Scikit‐learn package was used to create a predictive model using TYK2 JH2 inhibitors from the ChEMBL database with reported biological activity. Using this technique, estimated IC50 values for natural chemical libraries were determined. Notably, seven compounds with high predicted IC50 values were shown to have JAK inhibitory action. Following a thorough examination of docking scores, binding affinities, and binding modes, six compounds were chosen and submitted to molecular dynamics modeling. The findings demonstrated that these compounds have potential inhibitory efficacy, diverse scaffolds, and similar results to deucravacitinib.