DEELIG: A Deep Learning Approach to Predict Protein-Ligand Binding Affinity

Abstract

Protein-ligand binding prediction has extensive biological significance. Binding affinity helps in understanding the degree of protein-ligand interactions and is a useful measure in drug design. Protein-ligand docking using virtual screening and molecular dynamic simulations are required to predict the binding affinity of a ligand to its cognate receptor. Performing such analyses to cover the entire chemical space of small molecules requires intense computational power. Recent developments using deep learning have enabled us to make sense of massive amounts of complex data sets where the ability of the model to “learn” intrinsic patterns in a complex plane of data is the strength of the approach. Here, we have incorporated convolutional neural networks to find spatial relationships among data to help us predict affinity of binding of proteins in whole superfamilies toward a diverse set of ligands without the need of a docked pose or complex as user input. The models were trained and validated using a stringent methodology for feature extraction. Our model performs better in comparison to some existing methods used widely and is suitable for predictions on high-resolution protein crystal (⩽2.5 Å) and nonpeptide ligand as individual inputs. Our approach to network construction and training on protein-ligand data set prepared in-house has yielded significant insights. We have also tested DEELIG on few COVID-19 main protease-inhibitor complexes relevant to the current public health scenario. DEELIG-based predictions can be incorporated in existing databases including RSCB PDB, PDBMoad, and PDBbind in filling missing binding affinity data for protein-ligand complexes.

On the Frustration to Predict Binding Affinities from Protein–Ligand Structures with Deep Neural Networks

“…None of them significantly outperforms firstgeneration CNNs, most models presenting rather similar accuracies (Pearson correlation coefficient in the 0.80-0.85 range; root-mean square error around 1.2-1.3 pK unit) in predicting affinities for the PDBbind core set (Table S1) but significantly lower accuracies for true external test sets. 31,33,35 Despite the strong commitment of data scientists, we believe that drug discovery has not really benefited from the already described models for the major reasons that machine (deep) learning scoring functions still generalize poorly and are not readily applicable to virtual screening of large compound libraries. 32 This major discrepancy does not prevent computer scientists to propose novel deep learning models, almost on a monthly basis, usually focusing on the novelty of the deep neural network architecture but often omitting to answer three questions: (i) is the apparent performance biased by either the chosen descriptors, [43][44] or the protein-ligand training space?…”

“…For example, drug discovery would immediately benefit from key advances in this topic, by better triaging potentially interesting molecules among virtual screening hits , and proposing viable analogues in emerging ultra-large chemical spaces for hit to lead optimization. With the ever increasing amount of high-resolution experimentally determined protein–ligand structures, binding affinity prediction algorithms have switched from physics-based to empirical scoring functions, and in the last few years to machine learning and deep learning methods. , The latter category of descriptor-based scoring functions has notably led to numerous protein–ligand affinity models − (see a nonexhaustive list Table S1) notably because deep learning does not require explicit descriptor engineering and is ideally suited to find hidden nonlinear relationships between 3D protein–ligand structures and binding affinity. The first deep neural networks (DNNs) to predict binding affinities were convolutional neural networks (CNNs) reading a protein–ligand complex as an ensemble of grid-based voxels with multiple channels corresponding to pharmacophoric properties. ,, The CNN architecture is relatively inefficient from a computational point of view because most of the voxels do not carry any relevant information.…”

“…Many graph neural networks (GNNs), used as end-to-end standalone architecture, ,,,, in cascade or in combination with CNNs, have been described recently. None of them significantly outperforms first-generation CNNs, most models presenting rather similar accuracies (Pearson correlation coefficient in the 0.80–0.85 range; root-mean square error (RMSE) around 1.2–1.3 pK unit) in predicting affinities for the PDBbind core set (Table S1) but significantly lower accuracies for true external test sets. ,, …”

In Vitro Investigation of Binding Interactions between Albumin–Gliclazide Model and Typical Hypotensive Drugs

“…Figure 3 shows the double-logarithm plot (log([F 0 − F]/F) vs. log(Q)) at different temperatures, in which the slope equals n and the length of the intercept on the y-axis equals log K a ; the data are presented in Table 2 . Most of the binding studies available in the literature concern simple binary protein–ligand models [ 18 , 19 ]. However, a more complex approach is desired, since diabetic patients usually receive more than one medicine and the risk of drug–drug interactions should be considered [ 20 , 21 ].…”

PandoraRL: DQN and Graph Convolution based ligand pose learning for SARS-COV1 Mprotease

“…This has led to the use of machine learning and AI driven methods for predicting binding affinity and ligand conformations, and effective representation of the molecules becomes crucial. Graph based methods have gained popularity due to their ability to provide learnable feature vectors (fingerprints) of defined length even for molecules of varying size [1, 2, 6, 7, 25]. These methods represent molecules as 2D graphs.…”