Bayesian neural network with pretrained protein embedding enhances prediction accuracy of drug-protein interaction.

MOTIVATION: Characterizing drug-protein interactions (DPIs) is crucial to the high-throughput screening for drug discovery. The deep learning-based approaches have attracted attention because they can predict DPIs without human trial and error. However, because data labeling requires significant resources, the available protein data size is relatively small, which consequently decreases model performance. Here, we propose two methods to construct a deep learning framework that exhibits superior performance with a small labeled dataset. RESULTS: At first, we use transfer learning in encoding protein sequences with a pretrained model, which trains general sequence representations in an unsupervised manner. Second, we use a Bayesian neural network to make a robust model by estimating the data uncertainty. Our resulting model performs better than the previous baselines at predicting interactions between molecules and proteins. We also show that the quantified uncertainty from the Bayesian inference is related to confidence and can be used for screening DPI data points. AVAILABILITY AND IMPLEMENTATION: The code is available at https://github.com/QHwan/PretrainDPI. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

GCIceNet: a graph convolutional network for accurate classification of water phases.

Understanding the phases of water molecules based on local structure is essential for understanding their anomalous properties. However, due to complicated structural motifs formed via hydrogen bonds, conventional order parameters represent water molecules incompletely. In this paper, we develop GCIceNet, which automatically generates machine-based order parameters for classifying the phases of water molecules via supervised and unsupervised learning. The multiple graph convolutional layers in GCIceNet can learn topological information on the complex hydrogen bond networks. It shows a substantial improvement in accuracy for predicting the phase of water molecules in a bulk system and an ice/vapor interface system. A relative importance analysis shows that GCIceNet can capture the structural features of the given system hidden in the input data. Augmented with the vast amount of data provided by molecular dynamics simulations, GCIceNet is expected to serve as a powerful tool for the fields of glassy liquids and hydration layers around biomolecules.

Interfacial thermodynamics of spherical nanodroplets: molecular understanding of surface tension via a hydrogen bond network.

Surface tension plays a ubiquitous role in phase transitions including condensation or evaporation of atmospheric liquid droplets. In particular, understanding of interfacial thermodynamics of the critical nucleus of 1 nm scale is important for molecular characterization of the activation energy barrier of nucleation. Here, we investigate surface tension of spherical nanodroplets with both molecular dynamics and density functional theory and find that surface tension decreases appreciably below 1 nm radius, whose analytical expression is consistently derived from the classic Tolman's equation. In particular, the free energy analysis of nanodroplets shows that the change of surface tension originates dominantly from the configurational energy of interfacial molecules, which is evidenced by the increasingly disrupted hydrogen bond network as the droplet size decreases. Our result can be applied to the interface-related phenomena associated with molecular fluctuations such as biomolecule adsorption at the sub-nm scale where macroscopic thermodynamic quantities are ill-defined.

Multistage Transformation and Lattice Fluctuation at AgCl-Ag Interface.

Solid-state transformation between different materials is often accompanied by mechanical expansion and compression due to their volume change and structural evolution at interfaces. However, these two types of dynamics are usually difficult to monitor in the same time. In this work, we use in situ transmission electron microscopy to directly study the reduction transformation at the AgCl-Ag interface. Three stages of lattice fluctuations were identified and correlated to the structural evolution. During the steady state, a quasi-layered growth mode of Ag in both vertical and lateral directions were observed due to the confinement of AgCl lattices. The development of planar defects and depletion of AgCl are respectively associated with lattice compression and relaxation. Topography and structure of decomposing AgCl was further monitored by in situ scanning transmission electron microscopy. Silver species are suggested to originate from both the surface and the interior of AgCl, and be transported to the interface. Such mass transport may have enabled the steady state and lattice compression in this volume-shrinking transformation.

Viscometry of single nanoliter-volume droplets using dynamic force spectroscopy.

The viscometry of minute amounts of liquid has been in high demand as a novel tool for medical diagnosis and biological assays. Various microrheological techniques have shown the capability to handle small volumes. However, as the liquid volume decreases down to nanoliter scale, increasingly dominant surface effects complicate the measurement and analysis, which remain a challenge in microrheology. Here, we demonstrate an atomic force microscope-based platform that determines the viscosity of single sessile drops of 1 nanoliter Newtonian fluids. We circumvent interfacial effects by measuring the negative-valued shear elasticity, originating from the retarded fluidic response inside the drop. Our measurement is independent of the liquid-boundary effects, and thus is valid without a priori knowledge of surface tension or contact angle, and consistently holds at a 1 milliliter-scale volume. Importantly, while previous methods typically need a much larger 'unrecoverable' volume above 1 microliter, our simple platform uses only ∼1 nanoliter. Our results offer a quantitative and unambiguous methodology for viscosity measurements of extremely minute volumes of Newtonian liquids on the nanoliter scale.

Superwetting of TiO2 by light-induced water-layer growth via delocalized surface electrons.

Titania, which exhibits superwetting under light illumination, has been widely used as an ideal material for environmental solution such as self-cleaning, water-air purification, and antifogging. There have been various studies to understand such superhydrophilic conversion. The origin of superwetting has not been clarified in a unified mechanism yet, which requires direct experimental investigation of the dynamic processes of water-layer growth. We report in situ measurements of the growth rate and height of the photo-adsorbed water layers by tip-based dynamic force microscopy. For nanocrystalline anatase and rutile TiO2 we observe light-induced enhancement of the rate and height, which decrease after O2 annealing. The results lead us to confirm that the long-range attraction between water molecules and TiO2, which is mediated by delocalized electrons in the shallow traps associated with O2 vacancies, produces photo-adsorption of water on the surface. In addition, molecular dynamics simulations clearly show that such photo-adsorbed water is critical to the zero contact angle of a water droplet spreading on it. Therefore, we conclude that this "water wets water" mechanism acting on the photo-adsorbed water layers is responsible for the light-induced superwetting of TiO2. Similar mechanism may be applied for better understanding of the hydrophilic conversion of doped TiO2 or other photo-catalytic oxides.

Unified stress tensor of the hydration water layer.

We present the general stress tensor of the ubiquitous hydration water layer (HWL), based on the empirical hydration force, by combining the elasticity and hydrodynamics theories. The tapping and shear component of the tensor describe the elastic and damping properties of the HWL, respectively, in good agreement with experiments. In particular, a unified understanding of HWL dynamics provides the otherwise unavailable intrinsic parameters of the HWL, which offer additional but unexplored aspects to the supercooled liquidity of the confined HWL. Our results may allow deeper insight on systems where the HWL is critical.