Machine Learning-Guided Design of Novel Janus Kinase Inhibitors for Rheumatoid Arthritis Treatment

Abstract

Attenuation of autoimmune responses by inhibiting immune activation pathways has been a common strategy for treating autoimmune diseases. The inhibition of JAK2 protein has proven effective in a clinical setting for treating RA. However, most of the JAK inhibitors have limited isoform selectivity, resulting in broad immune suppression that elevates the risk of infection, malignancy, and cardiovascular events. These concerns promote the development of novel JAK inhibitors with improved potency and selectivity. Focusing on the JAK2 protein, we identified in-silico novel small molecules with higher binding affinity than the existing drugs through a computational screening process that combines the generation of a drug compound library, development of predicative machine learning model, and molecular dynamics-based modeling. From a generated library of 34,992 candidates, four compounds were identified with enhanced binding affinity relative to existing inhibitors. Notably, the top candidate, M-84, exhibited a binding free energy of - 47.1 kcal/mol, which is 13.4 kcal/mol lower than that of the strongest reference drug. An analysis of the interaction profiles revealed that the additional hydrogen bonding and salt-bridge interactions within the JAK2 binding pocket enabled by the additional sulfonyl and alkylammonium groups contributed to the enhanced binding affinity, suggesting that tailoring functional groups to the chemical environment of the binding pocket can guide the rational design of new inhibitors. Moreover, the computational pipeline presented here is readily transferable to the discovery of potent small-molecule inhibitors targeting other proteins.