PLAS-5k: Dataset of Protein-Ligand Affinities from Molecular Dynamics for Machine Learning Applications

Abstract

Computational methods and recently modern machine learning methods have played a key role in structure-based drug design. Though several benchmarking datasets are available for machine learning applications in virtual screening, accurate prediction of binding affinity for a protein-ligand complex remains a major challenge. New datasets that allow for the development of models for predicting binding affinities better than the state-of-the-art scoring functions are important. For the first time, we have developed a dataset, PLAS-5k comprised of 5000 protein-ligand complexes chosen from PDB database. The dataset consists of binding affinities along with energy components like electrostatic, van der Waals, polar and non-polar solvation energy calculated from molecular dynamics simulations using MMPBSA (Molecular Mechanics Poisson-Boltzmann Surface Area) method. The calculated binding affinities outperformed docking scores and showed a good correlation with the available experimental values. The availability of energy components may enable optimization of desired components during machine learning-based drug design. Further, OnionNet model has been retrained on PLAS-5k dataset and is provided as a baseline for the prediction of binding affinities.

Cordycepin Triphosphate as a Potential Modulator of Cellular Plasticity in Cancer via cAMP-Dependent Pathways: An In Silico Approach

“…However, our focus was not on determining the exact binding mode of these ligands but rather on supporting our hypothesis which held that the high structural and physicochemical similarities between COR (and its derivatives) and the endogenous metabolites derived from purine, could be extrapolated to comparable affinities and binding modes in their respective targets. Finally, it is noteworthy that docking-based methodologies have been reported to show a wide variation in their results when compared with experimental evidence [ 119 ]. For this reason, we conducted molecular docking studies with two additional tools different from the one used here, and the data can be found in Supplementary Material File S1 .…”

MISATO: machine learning dataset of protein–ligand complexes for structure-based drug discovery

“…Several databases are available that contain raw experimental structures of protein–ligand complexes, usually extracted from the PDB (for example, PDBbind 25 , bindingDB 26 , Binding MOAD 27 , Sperrylite 28 ). Only recently a database of MD-derived traces of protein–ligand structures was reported 29 , 30 . Despite these efforts, so far no AI model has been proposed that convincingly addresses the rational DD challenge in the way that AlphaFold2 answered the protein structure prediction problem 31 , 32 .…”

Synthesis and molecular docking studies of novel tricyclic and angular tetracyclic benzothiadiazines via sp3‐C‐Hactivation as potential colon cancer inhibitors

“…Prepared ligands and reference drugs were docked into the active sites (determined using Discovery studio 2020 software) of all colon cancer protein molecules using Pyrx 0.8 software with resolutions ranging from 1.90 to 2.30 Å within 90 × 75 × 60 cubic grid centers. A grid spacing of 1.00 Å was used for the calculation of the grid maps using the autogrid module of AutoDock tools, and for each ligand, a set of nine (9) independent runs were performed for all enzymes run against all ligands and reference drugs [44]. Previously docked heteroatoms, all water molecules, and other unwanted entities were removed, while polar hydrogen atoms were also added [45].…”