Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2020.04.20.050393

Deep learning model can predict water binding sites on the surface of proteins using limited-resolution data

Zaucha, J.·Softley, C. A.·Sattler, M.·Popowicz, G. M.
View original source ↗Export metadata ↓

Abstract

The surfaces of proteins are generally hydrophilic but there have been reports of sites that exhibit an exceptionally high affinity for individual water molecules. Not only do such molecules often fulfil critical biological functions, but also, they may alter the binding of newly designed drugs. In crystal structures, sites consistently occupied in each unit cell yield electron density clouds that represent water molecule presence. These are recorded in virtually all high-resolution structures obtained through X-ray diffraction. In this work, we utilized the wealth of data from the RCSB Protein Data Bank to train a residual deep learning model named hotWater to identify sites on the surface of proteins that are most likely to bind water, the so-called water hot spots. The model can be used to score existing water molecules from a PDB file to provide their ranking according to the predicted binding strength or to scan the surface of a protein to determine the most likely water hot-spots de novo. This is computationally much more efficient than currently used molecular dynamics simulations. Based on testing the model on three example proteins, which have been resolved using both high-resolution X-ray crystallography (providing accurate positions of trapped waters) as well as low-resolution X-ray diffraction, NMR or CryoEM (where structure refinement does not yield water positions), we were able to show that the hotWater method is able to recover in the "water-free" structures many water binding sites known from the high-resolution structures. A blind test on a newly solved protein structure with waters removed from the PDB also showed good prediction of the crystal water positions. This was compared to two known algorithms that use electron density and was shown to have higher recall at resolutions >2.6 [A]. We also show that the algorithm can be applied to novel proteins such as the RNA polymerase complex from SARS-CoV-2, which could be of use in drug discovery. The hotWater model is freely available at (https://pypi.org/project/hotWater/).

Source connections

Explore related subjects

bioinformaticsPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Zaucha, J., Softley, C. A., Sattler, M., Popowicz, G. M.. 2020-04-21. Deep learning model can predict water binding sites on the surface of proteins using limited-resolution data. https://doi.org/10.1101/2020.04.20.050393

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

RoDPAA: A Robust Dissimilarity-Based Principal Angle Analysis Framework for Multi-View Microbiome Data Integration

Multiview microbiome studies profile the same participants across body sites or molecular layers, but integration is hampered by sparse, compositional data, modest sample sizes and methods that recognize only exact sharing. We present RoDPAA (Robust Dissimilarity-based Principal Angle Analysis), a framework that represents each view through any data-appropriate dissimilarity, such as Bray--Curtis, without log-ratio transformation or likelihood assumptions. RoDPAA aligns view-specific signal subspaces through principal angles, captures complete and partial sharing, and tests by permutation how many components are shared, using only eigendecomposition and little tuning. In Gaussian simulations, RoDPAA matched JIVE and MOFA given true ranks, ran faster and alone recovered the shared dimension in every replicate. In two inflammatory bowel disease (IBD) cohorts with sparse, zero-inflated data, Bray--Curtis-based RoDPAA recovered shared structure that was stable to prevalence filtering, whereas log-ratio-based alternatives were not. Its leading microbiome--metabolome factors, learned without disease labels, were the only ones associated with IBD.

bioinformatics↗
bioRxiv2026

Deducing splicing events from transcript sets with splicelogic

Long-read RNA sequencing captures a greater proportion of full-length transcripts, enabling more accurate isoform identification, discovery of novel isoforms, and more precise differential transcript usage analysis. However, comparing transcripts, whether between conditions or between sets such as novel and annotated isoforms, yields complex isoform changes that are hard to manipulate, summarize, and visualize. This is compounded by vertical integration of some tools, where downstream analysis depends on the specific output of a particular upstream tool. To address this gap, we have developed splicelogic, an R/Bioconductor package that makes transcript-set comparisons biologically interpretable by turning them into discrete splicing events, such as exon skipping, alternative splice sites, intron retention, and alternative transcription start and end sites. We designed splicelogic to integrate easily with existing RNA-seq workflows and tools. The splicelogic functions are modular, readable, and extensible, making it easy for users to adapt core logic to their own custom scientific questions or novel splicing definitions. By obtaining interpretable splicing events from transcript changes, splicelogic enables downstream analyses, including annotation of RNA-binding protein recognition sites, alongside summary statistics and visualization. The splicelogic package is publicly available and distributed through Bioconductor.

bioinformatics↗
bioRxiv2026

Contrastive Learning Improves Generalization on M2OR Olfactory Binding Dataset

Olfaction-based biosensors stand to have profound impacts in fields such as medical diagnostics, national security, and drug discovery. A promising mode of vapor phase biodetection mimics volatile organic compound-olfactory receptor (VOC-OR) interactions with target-specific protein or peptide probes; however, lack of experimental coverage remains a core bottleneck. Current computational approaches aim to improve prediction of VOC-OR binding to reduce experimental burden, and current machine learning approaches fail to demonstrate a level of generalization that addresses current engineering challenges. Here, using the Molecular to Olfactory Dataset (M2OR) and Insect Olfactory Binding Protein Database (iOBPdb) datasets, data partitioning methods and their performance are evaluated on a neural network and contrastive learning framework on such VOC-OR binding datasets, varying: model training, architecture, and structure of the input data distribution. It was found that holding out certain chemical or proteomic motifs from the training data diminished model performance with a two layer neural network model. The model was able to generalize to state of the art performance (0.88 AUCROC) on every data split by applying a contrastive learning objective defined by shared molecular constituents and a stochastic sampler. This suggests the contrastive objective provides a strong inductive bias for neural networks to predict novel VOC-OR pairs. These findings highlight that model performance on pairwise biological interaction datasets can be inflated through identity leakage from constituent overlap, and emphasize the need for more robust evaluation protocols. Furthermore, we show contrastive regularization mitigates dataset degeneracy through enforcement of compatibility in the models latent geometry that may extend to other binder target-datasets where identity leakage confounds evaluation.

bioinformatics↗