Cathepsins belong to the family of cysteine proteases and are present in both the extracellular matrix and acidic lysosomes. They participate in apoptosis, antigen processing, bone resorption, and other enzymatic processes. Cathepsins are synthesized as inactive precursors (procathepsins) that contain a propeptide sterically blocking the active site. The enzymatic activity of (pro)cathepsins can be regulated by glycosaminoglycans (GAGs), which are linear, anionic, and sulfated carbohydrates composed of aminosugar/uronic acid dimeric units. By binding to allosteric sites on cathepsins’ surfaces, GAGs can induce conformational changes in the protein and thereby modulate its enzymatic activity. In this study, we examined the effect of GAG binding on procathepsin K flexibility. Using molecular docking, potential allosteric binding sites were identified and the most stable procathepsin K-GAG complexes were selected for detailed analysis. Furthermore, it was observed that procathepsin K can spontaneously undergo conformational changes at acidic pH, in contrast to neutral pH, which is consistent with experimental observations. In addition, different machine learning models, including linear regression, decision trees, and neural networks, were trained to predict binding free energies in (pro)cathepsin-GAG complexes. As input data, structural descriptors such as interatomic distances, hydrogen bond counts, solvent-accessible surface area (SASA), and electrostatic properties were employed. The most accurate predictions were achieved with a fully connected neural network of simple architecture and low dropout rate. Overall, our results provide novel insights into the structural mechanisms of GAG-mediated regulation of procathepsin K and demonstrate the utility of machine learning approaches as efficient tools for screening and prioritizing potential binders.
Krzysztof K. Bojarski (Sun,) studied this question.