arXiv subjects

Alexander Hudson

Publications and source records attributed to Alexander Hudson.

3 recordsLinked to original sources

arXiv2020

Transfer Learning for Protein Structure Classification at Low Resolution

Structure determination is key to understanding protein function at a molecular level. Whilst significant advances have been made in predicting structure and function from amino acid sequence, researchers must still rely on expensive, time-consuming analytical methods to visualise detailed protein conformation. In this study, we demonstrate that it is possible to make accurate ($\geq$80%) predictions of protein class and architecture from structures determined at low ($>$3A) resolution, using a deep convolutional neural network trained on high-resolution ($\leq$3A) structures represented as 2D matrices. Thus, we provide proof of concept for high-speed, low-cost protein structure classification at low resolution, and a basis for extension to prediction of function. We investigate the impact of the input representation on classification performance, showing that side-chain information may not be necessary for fine-grained structure predictions. Finally, we confirm that high-resolution, low-resolution and NMR-determined structures inhabit a common feature space, and thus provide a theoretical foundation for boosting with single-image super-resolution.

cs.CV↗

arXiv2018

On the nature of the glass transition in atomistic models of glass formers

We study the nature of the glass transition by cooling model atomistic glass formers at constant rate from a temperature above the onset of glassy dynamics to $T=0$. Motivated by the East model, a kinetically constrained lattice model with hierarchical relaxation, we make several predictions about the behavior of the supercooled liquid as it passes through the glass transition. We then compare those predictions to the results of our atomistic simulations. Consistent with our predictions, our results show that the relaxation time $τ$ of the material undergoes a crossover from super-Arrhenius to Arrhenius behavior at a cooling-rate-dependent glass transition temperature $T_\text{g}$. The slope of $\lnτ$ with respect to inverse temperature exhibits a peak near $T_\text{g}$ that grows more pronounced with slower cooling, matching our expectations qualitatively. Additionally, the limiting value of this slope at low temperature shows remarkable quantitative agreement with our predictions. Our results also show that the rate of short-time particle displacements deviates from the equilibrium linear scaling around $T_\text{g}$, asymptotically approaching a different linear scaling. To our surprise, these short-time displacements, the dynamic indicators of the underlying excitations responsible for structural relaxation, show no spatial correlations beyond a few particle diameters, both above and below $T_\text{g}$. This final result is contrary to our expectation, based on previous results for East model glasses formed by cooling, that inter-excitation correlations should emerge as the liquid vitrifies.

cond-mat.stat-mech↗

arXiv2015

Rough interfaces, accurate predictions: The necessity of capillary modes in a minimal model of nanoscale hydrophobic solvation

Modern theories of the hydrophobic effect highlight its dependence on length scale, emphasizing in particular the importance of interfaces that emerge in the vicinity of sizable hydrophobes. We recently showed that a faithful treatment of such nanoscale interfaces requires careful attention to the statistics of capillary waves, with significant quantitative implications for the calculation of solvation thermodynamics. Here we show that a coarse-grained lattice model in the spirit of those pioneered by Chandler and coworkers, when informed by this understanding, can capture a broad range of hydrophobic behaviors with striking accuracy. Specifically, we calculate probability distributions for microscopic density fluctuations that agree very well with results of atomistic simulations, even many standard deviations from the mean, and even for probe volumes in highly heterogeneous environments. This accuracy is achieved without adjustment of free parameters, as the model is fully specified by well-known properties of liquid water. As illustrative examples of its utility, we characterize the free energy profile for a solute crossing the air-water interface, and compute the thermodynamic cost of evacuating the space between extended nanoscale surfaces. Together, these calculations suggest that a highly reduced model for aqueous solvation can serve as the basis for efficient multiscale modeling of spatial organization driven by hydrophobic and interfacial forces.

cond-mat.stat-mech↗