FlexE ensemble docking approach to virtual screening for CDK2 inhibitors

Abstract

In spite of a proven potential and effectiveness of FlexE in docking flexible ligands into an ensemble of protein structures, FlexE has rarely been successful in virtual screening situations. In this study, we constructed cyclin-dependent kinase 2 (CDK2) ensemble structures which have exactly the same backbone conformations as 1AQ1 but differ only at the side chain torsion angles of the key amino acid (Lys33, Phe80, Lys89 and Asp145) residues: the torsion angles observed in the 17 CDK2 crystal structures were adapted to represent conformational flexibility. FlexE ensemble docking protocol then completely samples the full conformational fields generated by combination of torsions of the four amino acids in the ATP binding site. Virtual screening for CDK2 inhibitors by using the FlexE ensemble docking of a database composed of 48,703 inactives and 82 actives showed significant enrichment factor (EF ¼ 18.5), and successfully identified 71 actives among the top 132 ligands (53.8%) ranked by total energy scores. Moreover, total energy scoring followed by visual inspection filtered-off non-specific binders among the highly-ranked ligands to increase the ratio of actives-to-inactives to 71:13 at the top 5% of the virtual screening solutions.

Closing the side-chain gap in protein loop modeling

“…While each method returns a single model for the side-chain placement, our study shows that an ensemble of side-chain orientations generated by the collective use of multiple methods can often lead to better models, this is particularly true for Prime, ICM and Orchestrar. These results are consistent with results from ensemble docking experiments [22][23][24]. However, given multiple side-chain orientations for a loop, current scoring functions are unable to distinguish which side-chain orientation is best, therefore this work supports further development of scoring functions for side-chain optimization.…”

Understanding the challenges of protein flexibility in drug design

“…The interesting aspect of this approach, as opposed to 4D docking, is that new conformations are created during the search, thus increasing structural variation. Several studies have reported good results with this approach [130], [131]. However, it remains a discrete method, strongly influenced by the content of the ensemble.…”

Scoring Ensembles of Docked Protein:Ligand Interactions for Virtual Lead Optimization

“…A priori, it may appear that a Boltzmann weighting scheme in which the ensemble follows a probability distribution function would be most appropriate. However, the literature describes several popular methods for evaluating ensembles including averaging, − Boltzmann weighting, creating a composite grid or united description of the ensemble, , and decision trees or algorithms to determine the most essential member(s). − While much research on ensembles has been conducted, there is no single convention for evaluation, and to our knowledge the basis for successful methods is rarely investigated.…”

“…Many docking programs have algorithms for handling ligand flexibility; however, receptor flexibility is only marginally considered . In previous work, we ,,, and others ,− ,− have established that the use of receptor ensembles created using NMR, molecular dynamics simulations, or multiple experimentally determined models aids in representing a flexible receptor. In addition, we have shown that docking to an ensemble leads to increased ranking accuracy for virtual lead optimization and that the ensembles can be pruned a priori to include only the critical members that maintain a conserved binding core …”