Accurate Prediction of Hydration Sites of Proteins Using Energy Model With Atom Embedding

Water in drug design: pitfalls and good practices

INTRODUCTION

Structure-based drug design relies on optimizing drug-target interactions and blocking harmful pathophysiological events at the atomic level. Such events of the human body are modulated by water acting either as a medium or an individual partner in molecular interactions. A precise understanding of the modulatory mechanisms of water is essential for a successful drug design.

AREAS COVERED

The present review discusses different topographical and networking situations that result in radically different roles of water, a root of various pitfalls of drug design. The review surveys good practices for tackling the problems of determining water structure at atomic resolution. Techniques for quantifying the effects of bulk, networking, and individual water molecules on the stability of drug-target complexes are also discussed. The article is based on a literature search using the PubMed, Web of Science, and Google Scholar databases.

EXPERT OPINION

With advances in rapid computational algorithms and a better understanding of the physicochemical machinery of complex formation, theoretical approaches have resulted in elegant and cost-effective tools that fill the knowledge gaps left by the limited experimental methods. Overcoming the technical pitfalls of drug design, water transforms from a frustrating challenge into a handy tool for fine-tuning drug-target interactions.

Improvement in positional accuracy of neural-network predicted hydration sites of proteins by incorporating atomic details of water-protein interactions and site-searching algorithm

Visualization of hydration structures over the entire protein surface is necessary to understand why the aqueous environment is essential for protein folding and functions. However, it is still difficult for experiments. Recently, we developed a convolutional neural network (CNN) to predict the probability distribution of hydration water molecules over protein surfaces and in protein cavities. The deep network was optimized using solely the distribution patterns of protein atoms surrounding each hydration water molecule in high-resolution X-ray crystal structures and successfully provided probability distributions of hydration water molecules. Despite the effectiveness of the probability distribution, the positional differences of the predicted positions obtained from the local maxima as predicted sites remained inadequate in reproducing the hydration sites in the crystal structure models. In this work, we modified the deep network by subdividing atomic classes based on the electronic properties of atoms composing amino acids. In addition, the exclusion volumes of each protein atom and hydration water molecule were taken to predict the hydration sites from the probability distribution. These information on chemical properties of atoms leads to an improvement in positional prediction accuracy. We selected the best CNN from 47 CNNs constructed by systematically varying the number of channels and layers of neural networks. Here, we report the improvements in prediction accuracy by the reorganized CNN together with the details in the architecture, training data, and peak search algorithm.

SuperWater: Predicting Water Molecule Positions on Protein Structures by Generative AI

Water molecules play a significant role in maintaining protein structural stability and facilitating molecular interactions. Accurate prediction of water molecule positions around protein structures is essential for understanding their biological roles and has significant implications for protein engineering and drug discovery. Here, we introduce SuperWater, a novel generative AI framework that integrates a score-based diffusion model with equivariant graph neural networks to predict water molecule placements around proteins with high accuracy. SuperWater surpasses existing methods, delivering state-of-the-art performance in both crystal water coverage and prediction precision, achieving water localization within 0.3 ± 0.06 Å of experimentally validated positions. We demonstrate the capabilities of SuperWater through case studies involving protein hydration, protein-ligand binding, and protein-protein binding sites. This framework can be adapted for various applications, including structural biology, binding site prediction, multi-body docking, and water-mediated drug design.

HydraProt: A New Deep Learning Tool for Fast and Accurate Prediction of Water Molecule Positions for Protein Structures

Water molecules are integral to the structural stability of proteins and vital for facilitating molecular interactions. However, accurately predicting their precise position around protein structures remains a significant challenge, making it a vibrant research area. In this paper, we introduce HydraProt (deep Hydration of Proteins), a novel methodology for predicting precise positions of water molecule oxygen atoms around protein structures, leveraging two interconnected deep learning architectures: a 3D U-net and a Multi-Layer Perceptron (MLP). Our approach starts by introducing a coarse voxel-based representation of the protein, which allows for rapid sampling of candidate water positions via the 3D U-net. These water positions are then assessed by embedding the water-protein relationship in the Euclidean space by means of an MLP. Finally, a postprocessing step is applied to further refine the MLP predictions. HydraProt surpasses existing state-of-the-art approaches in terms of precision and recall and has been validated on large data sets of protein structures. Notably, our method offers rapid inference runtime and should constitute the method of choice for protein structure studies and drug discovery applications. Our pretrained models, data, and the source code required to reproduce these results are accessible at https://github.com/azamanos/HydraProt.

Ensemble structure of the N-terminal domain (1-267) of FUS in a biomolecular condensate

Solutions of some proteins phase separate into a condensed state of high protein concentration and a dispersed state of low concentration. Such behavior is observed in living cells for a number of RNA-binding proteins that feature intrinsically disordered domains. It is relevant for cell function via the formation of membraneless organelles and transcriptional condensates. On a basic level, the process can be studied in vitro on protein domains that are necessary and sufficient for liquid-liquid phase separation (LLPS). We have performed distance distribution measurements by electron paramagnetic resonance for 13 sections in an N-terminal domain (NTD) construct of the protein fused in sarcoma (FUS), consisting of the QGSY-rich domain and the RGG1 domain, in the denatured, dispersed, and condensed state. Using 10 distance distribution restraints for ensemble modeling and three such restraints for model validation, we have found that FUS NTD behaves as a random-coil polymer under good-solvent conditions in both the dispersed and condensed state. Conformation distribution in the biomolecular condensate is virtually indistinguishable from the one in an unrestrained ensemble, with the latter one being based on only residue-specific Ramachandran angle distributions. Over its whole length, FUS NTD is slightly more compact in the condensed than in the dispersed state, which is in line with the theory for random coils in good solvent proposed by de Gennes, Daoud, and Jannink. The estimated concentration in the condensate exceeds the overlap concentration resulting from this theory. The QGSY-rich domain is slightly more extended, slightly more hydrated, and has slightly higher propensity for LLPS than the RGG1 domain. Our results support previous suggestions that LLPS of FUS is driven by multiple transient nonspecific hydrogen bonding and π-sp2 interactions.

De novo prediction of explicit water molecule positions by a novel algorithm within the protein design software MUMBO

By mediating interatomic interactions, water molecules play a major role in protein-protein, protein-DNA and protein-ligand interfaces, significantly affecting affinity and specificity. This notwithstanding, explicit water molecules are usually not considered in protein design software because of high computational costs. To challenge this situation, we analyzed the binding characteristics of 60,000 waters from high resolution crystal structures and used the observed parameters to implement the prediction of water molecules in the protein design and side chain-packing software MUMBO. To reduce the complexity of the problem, we incorporated water molecules through the solvation of rotamer pairs instead of relying on solvated rotamer libraries. Our validation demonstrates the potential of our algorithm by achieving recovery rates of 67% for bridging water molecules and up to 86% for fully coordinated waters. The efficacy of our algorithm is highlighted further by the prediction of 3 different proteinligand complexes. Here, 91% of water-mediated interactions between protein and ligand are correctly predicted. These results suggest that the new algorithm could prove highly beneficial for structure-based protein design, particularly for the optimization of ligand-binding pockets or protein-protein interfaces.

The Advances and Limitations of the Determination and Applications of Water Structure in Molecular Engineering

Water is a key actor of various processes of nature and, therefore, molecular engineering has to take the structural and energetic consequences of hydration into account. While the present review focuses on the target-ligand interactions in drug design, with a focus on biomolecules, these methods and applications can be easily adapted to other fields of the molecular engineering of molecular complexes, including solid hydrates. The review starts with the problems and solutions of the determination of water structures. The experimental approaches and theoretical calculations are summarized, including conceptual classifications. The implementations and applications of water models are featured for the calculation of the binding thermodynamics and computational ligand docking. It is concluded that theoretical approaches not only reproduce or complete experimental water structures, but also provide key information on the contribution of individual water molecules and are indispensable tools in molecular engineering.

Prediction of Water Distributions and Displacement at Protein-Ligand Interfaces

The retention and displacement of water molecules during formation of ligand-protein interfaces play a major role in determining ligand binding. Understanding these effects requires a method for positioning of water molecules in the bound and unbound proteins and for defining water displacement upon ligand binding. We describe an algorithm for water placement and a calculation of ligand-driven water displacement in >9000 protein-ligand complexes. The algorithm predicts approximately 38% of experimental water positions within 1.0 Å and about 83% within 1.5 Å. We further show that the predicted water molecules can complete water networks not detected in crystallographic structures of the protein-ligand complexes. The algorithm was also applied to solvation of the corresponding unbound proteins, and this allowed calculation of water displacement upon ligand binding based on differences in the water network between the bound and unbound structures. We illustrate use of this approach through comparison of water displacement by structurally related ligands at the same binding site. This method for evaluation of water displacement upon ligand binding may be of value for prediction of the effects of ligand modification in drug design.