Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets

Abstract

Deep generative models have achieved tremendous success in designing novel drug molecules in recent years. A new thread of works have shown the great potential in advancing the specificity and success rate of in silico drug design by considering the structure of protein pockets. This setting posts fundamental computational challenges in sampling new chemical compounds that could satisfy multiple geometrical constraints imposed by pockets. Previous sampling algorithms either sample in the graph space or only consider the 3D coordinates of atoms while ignoring other detailed chemical structures such as bond types and functional groups. To address the challenge, we develop Pocket2Mol, an E(3)-equivariant generative network composed of two modules: 1) a new graph neural network capturing both spatial and bonding relationships between atoms of the binding pockets and 2) a new efficient algorithm which samples new drug candidates conditioned on the pocket representations from a tractable distribution without relying on MCMC. Experimental results demonstrate that molecules sampled from Pocket2Mol achieve significantly better binding affinity and other drug properties such as druglikeness and synthetic accessibility.

De Novo Molecule Design Using Molecular Generative Models Constrained by Ligand–Protein Interactions

“…Their middle core structures are quite different to that of the seed compound. The similarity between the generated compounds and the active 38 Their model first estimates the probability density of atom's occurrences in the 3D space. Then, atoms are sampled sequentially from the learned distribution until there is no room for new atoms.…”

“…34−37 For example, Peng et al proposed a 3D generative model, which first predict the probability of grid points being occupied by an atom of a particular chemical element, then generate a diverse set of molecules following an auto-regressive sampling algorithm. 38 However, most of these 3D models are facing the challenge to generate high-quality and diverse drug-scale molecules. Recently, Xu et al built RNN generative models constrained by two sets of descriptors, which characterize the 3D information of protein binding pockets.…”

DiffSDS: A language diffusion model for protein backbone inpainting under geometric conditions and constraints

“…Molecule Conformation Generation G-SchNet (Gebauer et al, 2019) -PyTorch CVGAE (Mansimov et al, 2019) 2D-graph TF GraphDG (Simm et al, 2020b) 2D-graph PyTorch MolGym (Simm et al, 2020a) -PyTorch ConfGF (Shi et al, 2021) 2D-graph PyTorch ConfVAE (Xu et al, 2021) 2D-graph PyTorch DGSM 2D-graph -CGCF (Xu et al, 2020) 2D-graph PyTorch GeoMol (Ganea et al, 2021a) 2D-graph PyTorch G-SphereNet (Luo & Ji, 2021) -PyTorch GeoDiff (Xu et al, 2022) 2D-graph PyTorch EDM (Hoogeboom et al, 2022) 2D-graph PyTorch TorsionDiff (Jing et al, 2022) 2D-graph PyTorch DMCG (Zhu et al, 2022) 2D-graph PyTorch De novo Molecule Design DeLinker (Imrie et al, 2020) Protein Pocket 3D-fragments TF 3DMolNet (Nesterov et al, 2020) 3D-geometry -cG-SchNet (Gebauer et al, 2022) 3D-geometry PyTorch Luo's model (Luo et al, 2022) Protein Pocket PyTorch LiGAN (Ragoza et al, 2022) Protein Pocket PyTorch Bridge (Wu et al, 2022b) Physical prior -MDM (Huang et al, 2022a) 2D-graph Properties -Pocket2Mol (Peng et al, 2022) Protein Pocket PyTorch 3DLinkcer (Huang et al, 2022b) 3D-fragments PyTorch CGVAE (Wang et al, 2022b) Coarse Topology PyTorch GraphBP (Liu et al, 2022b) Protein Pocket PyTorch…”

“…A.1. Related works 3D Molecule Generation Generating 3D molecules to explore the local minima of the energy function (Conformation Generation) (Gebauer et al, 2019;Simm et al, 2020b;a;Shi et al, 2021;Xu et al, 2021;Xu et al, 2020;Ganea et al, 2021a;Xu et al, 2022;Hoogeboom et al, 2022;Jing et al, 2022;Zhu et al, 2022) or discover potential drug molecules binding to targeted proteins (3D Drug Design) (Imrie et al, 2020;Nesterov et al, 2020;Luo et al, 2022;Ragoza et al, 2022;Wu et al, 2022b;Huang et al, 2022a;Peng et al, 2022;Huang et al, 2022b;Wang et al, 2022b;Liu et al, 2022b) have attracted extensive attention in recent years. Compared to conformation generation that aims to predict the set of favourable conformers from the molecular graph, 3D Drug Design is more challenging in two aspects: (1) both conformation and molecule graph need to be generated, and (2) the generated molecules should satisfy multiple constraints, such as physical prior and protein-ligand binding affinity.…”

Application of Computational Biology and Artificial Intelligence in Drug Design

“…The Pocket2Mol has learned a probability distribution of atoms and bond types inside the pocked based on exiting atoms by adopting an auto-regression strategy, and used a graph neural network to capture features of atoms in binding pockets. For new drug sampling, this research considers the structures and geometrical constraints of protein pockets in drug design [ 255 ]. Another 3D generative model applied auto-regressive for novel molecule sampling can be found in study [ 240 ], similarly, it also used a neural network architecture to learn probability distribution of occurrences of atoms.…”