A Physics-Guided Neural Network for Predicting Protein–Ligand Binding Free Energy: From Host–Guest Systems to the PDBbind Database

Computer-assisted discovery and evaluation of potential ribosomal protein S6 kinase beta 2 inhibitors

S6K2 is an important protein in mTOR signaling pathway and cancer. To identify potential S6K2 inhibitors for mTOR pathway treatment, a virtual screening of 1,575,957 active molecules was performed using PLANET, AutoDock GPU, and AutoDock Vina, with their classification abilities compared. The MM/PB(GB)SA method was used to identify four compounds with the strongest binding energies. These compounds were further investigated using molecular dynamics (MD) simulations to understand the properties of the S6K2/ligand complex. Due to a lack of available 3D structures of S6K2, OmegaFold served as a reliable 3D predictive model with higher evaluation scores in SAVES v6.0 than AlphaFold, AlphaFold2, and RoseTTAFold2. The 150 ns MD simulation revealed that the S6K2 structure in aqueous solvation experienced compression during conformational relaxation and encountered potential energy traps of about 19.6 kJ mol-1. The virtual screening results indicated that Lys75 and Lys99 in S6K2 are key binding sites in the binding cavity. Additionally, MD simulations revealed that the ligands remained attached to the activation cavity of S6K2. Among the compounds, compound 1 induced restrictive dissociation of S6K2 in the presence of a flexible region, compound 8 achieved strong stability through hydrogen bonding with Lys99, compound 9 caused S6K2 tightening, and the binding of compound 16 was heavily influenced by hydrophobic interactions. This study suggests that these four potential inhibitors with different mechanisms of action could provide potential therapeutic options.

Calculation of protein-ligand binding entropies using a rule-based molecular fingerprint

The use of fast in silico prediction methods for protein-ligand binding free energies holds significant promise for the initial phases of drug development. Numerous traditional physics-based models (e.g., implicit solvent models), however, tend to either neglect or heavily approximate entropic contributions to binding due to their computational complexity. Consequently, such methods often yield imprecise assessments of binding strength. Machine learning models provide accurate predictions and can often outperform physics-based models. They, however, are often prone to overfitting, and the interpretation of their results can be difficult. Physics-guided machine learning models combine the consistency of physics-based models with the accuracy of modern data-driven algorithms. This work integrates physics-based model conformational entropies into a graph convolutional network. We introduce a new neural network architecture (a rule-based graph convolutional network) that generates molecular fingerprints according to predefined rules specifically optimized for binding free energy calculations. Our results on 100 small host-guest systems demonstrate significant improvements in convergence and preventing overfitting. We additionally demonstrate the transferability of our proposed hybrid model by training it on the aforementioned host-guest systems and then testing it on six unrelated protein-ligand systems. Our new model shows little difference in training set accuracy compared to a previous model but an order-of-magnitude improvement in test set accuracy. Finally, we show how the results of our hybrid model can be interpreted in a straightforward fashion.

Improving the Accuracy of Physics-Based Hydration-Free Energy Predictions by Machine Learning the Remaining Error Relative to the Experiment

The accuracy of computational models of water is key to atomistic simulations of biomolecules. We propose a computationally efficient way to improve the accuracy of the prediction of hydration-free energies (HFEs) of small molecules: the remaining errors of the physics-based models relative to the experiment are predicted and mitigated by machine learning (ML) as a postprocessing step. Specifically, the trained graph convolutional neural network attempts to identify the "blind spots" in the physics-based model predictions, where the complex physics of aqueous solvation is poorly accounted for, and partially corrects for them. The strategy is explored for five classical solvent models representing various accuracy/speed trade-offs, from the fast analytical generalized Born (GB) to the popular TIP3P explicit solvent model; experimental HFEs of small neutral molecules from the FreeSolv set are used for the training and testing. For all of the models, the ML correction reduces the resulting root-mean-square error relative to the experiment for HFEs of small molecules, without significant overfitting and with negligible computational overhead. For example, on the test set, the relative accuracy improvement is 47% for the fast analytical GB, making it, after the ML correction, almost as accurate as uncorrected TIP3P. For the TIP3P model, the accuracy improvement is about 39%, bringing the ML-corrected model's accuracy below the 1 kcal/mol threshold. In general, the relative benefit of the ML corrections is smaller for more accurate physics-based models, reaching the lower limit of about 20% relative accuracy gain compared with that of the physics-based treatment alone. The proposed strategy of using ML to learn the remaining error of physics-based models offers a distinct advantage over training ML alone directly on reference HFEs: it preserves the correct overall trend, even well outside of the training set.

Physics-Guided Deep Generative Model for New Ligand Discovery

Structure-based drug discovery aims to identify small molecules that can attach to a specific target protein and change its functionality. Recently, deep learning has shown great promise in generating drug-like molecules with specific biochemical features and conditioned with structural features. However, they usually fail to incorporate an essential factor: the underlying physics which guides molecular formation and binding in real-world scenarios. In this work, we describe a physics-guided deep generative model for new ligand discovery, conditioned not only on the binding site but also on physics-based features that describe the binding mechanism between a receptor and a ligand. The proposed hybrid model has been tested on large protein-ligand complexes and small host-guest systems. Using the top-N methodology, on average more than 75% of the generated structures by our hybrid model were stronger binders than the original reference ligand. All of them had higher ΔGbind (affinity) values than the ones generated by the previous state-of-the-art method by an average margin of 1.88 kcal/mol. The visualization of the top-5 ligands generated by the proposed physics-guided model and the reference deep learning model demonstrate more feasible conformations and orientations by the former. The future directions include training and testing the hybrid model on larger datasets, adding more relevant physics-based features, and interpreting the deep learning outcomes from biophysical perspectives.

Editorial: Special Issue “Protein Modeling and Simulation: Selected Articles from the Computational Structural Bioinformatics Workshop 2021”

Computational structural biology has demonstrated a key role in improving human health [...]