FlexAID: Revisiting Docking on Non-Native-Complex Structures

Abstract

Small-molecule protein docking is an essential tool in drug design and to understand molecular recognition. In the present work we introduce FlexAID, a small-molecule docking algorithm that accounts for target side-chain flexibility and utilizes a soft scoring function, i.e. one that is not highly dependent on specific geometric criteria, based on surface complementarity. The pairwise energy parameters were derived from a large dataset of true positive poses and negative decoys from the PDBbind database through an iterative process using Monte Carlo simulations. The prediction of binding poses is tested using the widely used Astex dataset as well as the HAP2 dataset, while performance in virtual screening is evaluated using a subset of the DUD dataset. We compare FlexAID to AutoDock Vina, FlexX, and rDock in an extensive number of scenarios to understand the strengths and limitations of the different programs as well as to reported results for Glide, GOLD, and DOCK6 where applicable. The most relevant among these scenarios is that of docking on flexible non-native-complex structures where as is the case in reality, the target conformation in the bound form is not known a priori. We demonstrate that FlexAID, unlike other programs, is robust against increasing structural variability. FlexAID obtains equivalent sampling success as GOLD and performs better than AutoDock Vina or FlexX in all scenarios against non-native-complex structures. FlexAID is better than rDock when there is at least one critical side-chain movement required upon ligand binding. In virtual screening, FlexAID results are lower on average than those of AutoDock Vina and rDock. The higher accuracy in flexible targets where critical movements are required, intuitive PyMOL-integrated graphical user interface and free source code as well as precompiled executables for Windows, Linux, and Mac OS make FlexAID a welcome addition to the arsenal of existing small-molecule protein docking methods.

Learning a force field from small-molecule crystal lattice predictions enables consistent sub-Angstrom protein-ligand docking

“…On this set, RosettaGenFF incorporated into GALigandDock achieved a median RMSD of 0.86 Å with success rates of 52/74% (using the criteria of ligand RMSD within 1/2 Å, respectively). This is an over 10% improvement in success rate over any previously reported study reported to date on the set 27,33,[48][49][50][51] ( Fig 3a ).…”

“…Incorporating receptor flexibility improves cross-docking. a) Success rates in cross-docking benchmark for various methods 27,33,[48][49][50][51] tested on Astex non-native set 33 . Blue and red bars represent results from docking runs with and without receptor flexibility, respectively; solid and patterned bars show results by two criteria, ligand RMSD < 2 Å and < 1 Å, respectively.…”

Large-scale detection of drug off-targets: hypotheses for drug repurposing and understanding side-effects

“…Docking simulations for targets with Z 3 were performed using FlexAID [30]. FlexAID is a probabilistic genetic-algorithm based method.…”

“…First, the new targets predicted could represent drug repurposing avenues and, second, they could be used to explain known side-effects of the drugs. For the most significant predictions, molecular docking simulations were performed using the FlexAID algorithm [30] to determine the potential docking pose of the drug in the potential cross-reactivity target. Poses of the ligand obtained by superimposing the drug bound target on the predicted cross-reactivity target using MIF similarities were compared to the pose obtained by the docking algorithm allowing to rationalize the prediction by looking at potential interactions in the target binding-site.…”

DUBS: A Framework for Developing Directory of Useful Benchmarking Sets for Virtual Screening

“…4 The virtual screening community has developed several benchmarking datasets to evaluate how well different docking methods perform at these vital tasks. [5][6][7][8][9][10][11][12][13] One of the popular benchmarking set is the Astex Diverse Benchmark 5 that is used to evaluate how well a given methodology can reproduce the crystal pose of a small molecule ligand in the bound (holo) form of a protein . Similarly, the PDBBind 14 and related CASF 11,15 datasets assess the ability of a docking method to produce and select a crystal-like pose for the small molecule ligand, and how well the methodology can rank the binding affinity of the small molecule.…”