How good are AlphaFold models for docking-based virtual screening?

Abstract

A crucial component in structure-based drug discovery is the availability of high-quality three-dimensional structures of the protein target. In molecular docking, whenever experimental structures were not available, in silico structural homology modeling has been the method of choice. However, using computationally predicted structures adds a further degree of uncertainty to the docking process. Recently, AlphaFold (AF), an artificial intelligence-based modeling tool, has shown impressive results in terms of model accuracy within the field of ab initio protein structure prediction. This outstanding success prompted us to evaluate how accurate AF models are from the perspective of docking-based drug discovery. We compared the performance of AF models in high-throughput docking (HTD) to their corresponding experimental PDB structures using a benchmark set of 16 targets spanning different protein families and binding site properties. Four docking programs and two consensus techniques were used to evaluate the HTD performance. The AF models showed consistently worse performance than their corresponding PDB structures, with zero enrichment factor values in several cases. While AlphaFold shows a remarkable ability to predict protein architecture and binding site anatomy, we conclude that this is not enough to guarantee that AF models can be reliably used for HTD purposes. Moreover, we show that very small variations at the side chain level of essential ligand-binding residues have a large impact on the outcome of HTD, what suggests that post-modeling refinement strategies might be key to increase the chance of success of AF models in prospective HTD campaigns.

Assessment of AlphaFold structures and optimization methods for virtual screening

“…In the test of 23 selected targets, the virtual screening effect of protein structures in DUD-E database is slightly better than that of AlphaFold structure, and AlphaFold structures are somewhat better than that of Apo structure. Our result demonstrate that AlphaFold exhibits a relative worse performance compared to Holo stuctures for VS, which is consistent with recent findings when comparing with Holo PDB structures 8 . Our results have guiding significance for the selection of protein structures in virtual screening.…”

“…These significant findings reveal practical applications of AlphaFold in predicting protein structure when multiple sequence alignment is unavailable. Although AlphaFold could produce protein structure information with high-accuracy, there is a need to better harness AlphaFold for VS 7,8 . We optimize the AlphaFold structure by molecular dynamics in order to improve the virtual screening.…”

“…Researchers find that rescoring our docking poses with three machine learning-based scoring functions increased the accuracy of predictions 7 . Furthermore, post-modeling refinement strategies may be helpful to improve the chances of success 8 .…”

“…Wong et al evaluated the effectiveness of AlphaFold protein structure to predict the binding affinity of 296 E. coli proteins and 218 small molecules with known antibacterial activity, and compared the results to the experimental structure of 12 proteins 6 . Scardino et al assessed the performance of the structure in the PDB database and AlphaFold structure in virtual screening using four docking software and ECR (Exponential Consensus Ranking, based on ranking) on 16 protein targets, and the consensus score of PRC (Pose/Ranking Consensus, based on docking pose and ranking) 7 . Zhang examined the efficiency of Holo, Apo and AlphaFold structures in virtual screening for 28 targets et al with Glide molecular docking method 8 .…”

Benchmarking Refined and Unrefined AlphaFold2 Structures for Hit Discovery

“…Cavasotto et al conducted a retrospective study to compare the docking performance between AF2 models and published crystal structures. 19 Docking into the AF2 structures produced lower enrichment than docking into the experimental structures. The authors suggest that modest refinement of the AF2 structures to address errors in side chain placement may be necessary to obtain an optimal docking performance.…”