The mechanisms of action of ligands competing for the Colchicine Binding Site (CBS) of the α,β-Tubulin are non-standard compared to the commonly witnessed ligand-induced inhibition of proteins. This is because their potencies are not solely judged by the binding affinity itself, but also by their capacity to bias the conformational states of the dimer. Regarding the latter requirement, it is observed that ligands competing for the same pocket that binds colchicine exhibit divergence in potential clinical outcomes. Molecular dynamics-based ∼5.2 µs sampling of α,β-Tubulin complexed with four different ligands has revealed that each ligand has its customized way of influencing the complex. Primarily, it is the proportion of twisting and/or bending characteristic of modes of the intrinsic dynamics which is revealed to be 'fundamental' to tune this variation in the mechanism. The milder influence of 'bending' makes a ligand (TUB092), better classifiable under the group of vascular disrupting agents (VDAs), which are phenotypically effective on cytoskeletons; whereas a stronger impact of 'bending' makes the classical ligand Colchicine (COL) a better Anti-Mitotic agent (AMA). Two other ligands BAL27862 (2RR) and Nocodazole (NZO) fall in the intermediate zone as they fail to explicitly induce bending modes. Random Forest Classification method and K-means Clustering is applied to reveal the efficiency of Machine Learning methods in classifying the Tubulin conformations according to their ligand-specific perturbations and to highlight the significance of specific amino acid residues, mostly positioned in the α-β and β-β interfaces involved in the mechanism. These key residues responsible to yield discriminative actions of the ligands are likely to be highly useful in future endeavours to design more precise inhibitors.
Keywords: Machine learning; Microtubules; Molecular dynamics; α; β-Tubulin.
Copyright © 2023 Elsevier Ltd. All rights reserved.