DeepDrug3D: Classification of ligand-binding pockets in proteins with a convolutional neural network

Abstract

Comprehensive characterization of ligand-binding sites is invaluable to infer molecular functions of hypothetical proteins, trace evolutionary relationships between proteins, engineer enzymes to achieve a desired substrate specificity, and develop drugs with improved selectivity profiles. These research efforts pose significant challenges owing to the fact that similar pockets are commonly observed across different folds, leading to the high degree of promiscuity of ligand-protein interactions at the system-level. On that account, novel algorithms to accurately classify binding sites are needed. Deep learning is attracting a significant attention due to its successful applications in a wide range of disciplines. In this communication, we present DeepDrug3D, a new approach to characterize and classify binding pockets in proteins with deep learning. It employs a state-of-the-art convolutional neural network in which biomolecular structures are represented as voxels assigned interaction energy-based attributes. The current implementation of DeepDrug3D, trained to detect and classify nucleotide- and heme-binding sites, not only achieves a high accuracy of 95%, but also has the ability to generalize to unseen data as demonstrated for steroid-binding proteins and peptidase enzymes. Interestingly, the analysis of strongly discriminative regions of binding pockets reveals that this high classification accuracy arises from learning the patterns of specific molecular interactions, such as hydrogen bonds, aromatic and hydrophobic contacts. DeepDrug3D is available as an open-source program at https://github.com/pulimeng/DeepDrug3D with the accompanying TOUGH-C1 benchmarking dataset accessible from https://osf.io/enz69/ .

Integrated Computational Approaches and Tools for Allosteric Drug Discovery

“…Pockets/cavities present in proteins might be functionally important and have been recognized as conventional sites for ligand binding [189]. For those targets that are not well characterized for such pockets, the identification of these sites is thus a starting point for potential allosteric-site prediction and structure-based drug design, with the effects of ligand binding confirming allostery [190].…”

DeepBindPoc: a deep learning method to rank ligand binding pockets using molecular vector representation

“…A recently developed P2Rank, a machine learning based tool, demonstrates a strong prediction ability for protein pocket (Krivák & Hoksza, 2018). There are several deep learning-based methods to identify native pockets (Jiménez et al, 2017;Pu et al, 2019). However, most current available methods (Saberi Fathi & Tuszynski, 2014;Jiménez et al, 2017;Krivák & Hoksza, 2018;Pu et al, 2019) have not incorporated ligand information in pocket identification, indicating these methods would have serious limitations in pocket which induces changes in protein structure upon ligand binding.…”

“…There are several deep learning-based methods to identify native pockets (Jiménez et al, 2017;Pu et al, 2019). However, most current available methods (Saberi Fathi & Tuszynski, 2014;Jiménez et al, 2017;Krivák & Hoksza, 2018;Pu et al, 2019) have not incorporated ligand information in pocket identification, indicating these methods would have serious limitations in pocket which induces changes in protein structure upon ligand binding. Some machine learning models have an over fitting problem which only performs well when a test case is close to the training data but fail to predict better on additional independent data (Ursenbach et al, 2019).…”

Defining a Global Map of Functional Group Based 3D Ligand-binding Motifs

“…With the rapid accumulation of experimentally determined structures of protein-ligand complexes, it became possible for large-scale identification of conserved 3D binding motifs using computational approaches (Kinjo and Nakamura, 2009;Ribeiro et al, 2020). Current methods based on structural comparison or alignment of protein pockets have identified many well-defined 3D motifs that are conserved across different protein pockets and widely used for protein function annotation, pockets classification and ligand-binding prediction (Gao and Skolnick, 2013;Hoffmann et al, 2010;Hwang et al, 2017;Pires et al, 2013;Pu et al, 2019;Yeturu and Chandra, 2008). However, these ligand-based 3D binding patterns are not applicable to large fraction of ligands especially small molecular drugs due to the lacking of reference 3D protein-ligand structures.…”