Search bioRxiv⌕ Search

bioRxiv · 10.1101/2022.04.18.488616

gr Predictor: a Deep-Learning Model for Predicting the Hydration Structures around Proteins

Abstract

Among the factors affecting biological processes such as protein folding and ligand binding, hydration, which is represented by a three-dimensional water-site-distribution-function around the protein, is crucial. The typical methods for computing the distribution functions, including molecular dynamics simulations and the three-dimensional reference interaction site model (3D-RISM) theory, require a long computation time from hours to tens of hours. Here, we propose a deep-learning model rapidly estimating the distribution functions around proteins obtained by the 3D-RISM theory from the protein 3D structure. The distribution functions predicted using our deep-learning model are in good agreement with those obtained by the 3D-RISM theory. Particularly, the coefficient of determination between the distribution function obtained by the deep-learning model and that obtained using the 3D-RISM theory is approximately 0.98. Furthermore, using a graphics processing unit (GPU), the calculation by the deep learning model is completed in less than one minute, more than 2 orders of magnitude faster than the calculation time of 3D-RISM theory. Therefore, our deep learning model provides a practical and efficient way to calculate the three-dimensional water-site-distribution-functions. The program called "gr Predictor" is available under the GNU General Public License from https://github.com/YoshidomeGroup-Hydration/gr-predictor. Table of Contents graphic O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=55 SRC="FIGDIR/small/488616v1_ufig1.gif" ALT="Figure 1"> View larger version (14K): org.highwire.dtl.DTLVardef@108b3c7org.highwire.dtl.DTLVardef@52ec18org.highwire.dtl.DTLVardef@a388borg.highwire.dtl.DTLVardef@10654ad_HPS_FORMAT_FIGEXP M_FIG C_FIG

Source connections

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Kawama, K., Fukushima, Y., Ikeguchi, M., Ohta, M., Yoshidome, T.. 2022-04-18. gr Predictor: a Deep-Learning Model for Predicting the Hydration Structures around Proteins. https://doi.org/10.1101/2022.04.18.488616

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

KEEP EXPLORING

Related preprints

Lipid Scrambling Amplifies Plasma Membrane Undulations via Anti-Registered Domains Stabilized by Long-Chain Sphingomyelin

Lipid scrambling, which dissipates the transbilayer lipid asymmetry of plasma membranes (PMs), plays an essential role in various large-scale membrane remodeling processes. Although lipid domains have been implicated in these processes, the underlying mechanisms remain poorly understood, owing to the lack of a molecular-level picture of how lipid scrambling reorganizes these domains. Here, using coarse-grained molecular dynamics simulations of asymmetric and fully scrambled PM models, we show that lipid scrambling markedly enhances membrane undulations. We identify two key features of the domain organization underlying this enhancement in the scrambled PM: domain anti-registration, which couples to membrane curvature through a mismatch in preferred curvature between the two leaflets, and scrambling-induced domain enlargement, which permits this curvature to extend over large length scales. Simulations with different sphingomyelin (SM) compositions demonstrate that long-chain SMs stabilize anti-registered domains and the associated membrane curvature. Together, these results suggest a physical mechanism by which lipid scrambling primes the PM for remodeling through the reorganization of lipid domains, with long-chain SMs playing a key role. We also find that the interleaflet domain organization of the scrambled PM model depends strongly on system size, shifting from anti-registration to registration as the simulation box is reduced, and we discuss the origin of this finite-size effect.

biophysics↗

Evaluating speech-like distortion product otoacoustic emissions: a comparison of time- and frequency-domain metrics across three datasets

Speech-like distortion-product otoacoustic emissions (DPOAEs) allow cochlear activity to be measured during the perception of natural, continuous speech and have been used to study attentional effects in the cochlea. The technique is still young, however, and existing studies differ in stimulus design and in the metric used to quantify the emissions, which makes their results difficult to compare. Here, we reanalysed three existing datasets (106 participants, 14 stimulus types) with a common pipeline. We compared six metrics: the band power of the recording, the magnitude-squared coherence (MSC) between the recording and a DPOAE template, and four metrics derived from their complex cross-correlation, namely peak magnitude, signal-to-noise ratio (SNR), peak area and peak width. A separability index quantified how well each metric separated in-ear recordings from out-of-ear control recordings without emissions. Band power, MSC and SNR separated best, with mean separability indices of about 0.9, whereas the peak magnitude performed worst and was inverted for half of the stimulus types. Since band power does not require a DPOAE template, it serves as a sanity check of the setup rather than as evidence for an emission. The MSC was insensitive to the latency of the emission but well suited for comparisons across stimuli, whereas the cross-correlation additionally provided the latency. The emissions differed significantly between the stimulus types of the three datasets, and those below about 700 Hz were measured least reliably. We derive practical recommendations for the analysis and stimulus design of future experiments.

biophysics↗

Overlap of competing voices may mask attentional effects on speech-like distortion product otoacoustic emissions

Whether selective auditory attention modulates the cochlear amplifier through the medial olivocochlear system remains debated. Using speech-like distortion-product otoacoustic emissions (DPOAEs), elicited by waveforms that follow the harmonics of natural speech, we previously found that selectively attending one of two competing talkers reduced the emissions related to the resolved harmonics of the female voice. Here, we attempted to reproduce this effect in a design that separated different frequency regimes and harmonic resolvability. Forty-four participants attended one of two competing talkers, a female and a male voice, presented to one ear, while speech-like DPOAEs related to both voices were recorded from the other ear. Stimuli were derived from resolved and unresolved harmonics in a low-frequency (up to 1.5 kHz) and a transitional (1.5 to 4 kHz) regime, which required pitch-shifting both voices so that their fundamental frequencies were much closer to each other than in the original study. Speech-like DPOAEs emerged for most stimulus types, although they were weak for some and absent for one. Their amplitude, however, did not depend on attention for any stimulus type. Further time-domain metrics showed only isolated attentional effects of opposite direction for two stimulus types, and a frequency-domain metric showed none. A cochlear excitation model suggests that, due to the closely spaced fundamental frequencies of the two voices, the emissions most comparable to the original study presumably arose at cochlear locations shared by both voices, where attention would modulate both voices alike.

biophysics↗