Dynamic weighting in Monte Carlo and optimization

Increasing the Sampling Efficiency of Protein Conformational Change by Combining a Modified Replica Exchange Molecular Dynamics and Normal Mode Analysis

Understanding conformational change at an atomic level is significant when determining a protein functional mechanism. Replica exchange molecular dynamics (REMD) is a widely used enhanced sampling method to explore protein conformational space. However, REMD with an explicit solvent model requires huge computational resources, immensely limiting its application. In this study, a variation of parallel tempering metadynamics (PTMetaD) with the omission of solvent-solvent interactions in exchange attempts and the use of low-frequency modes calculated by normal-mode analysis (NMA) as collective variables (CVs), namely ossPTMetaD, is proposed with the aim to accelerate MD simulations simultaneously in temperature and geometrical spaces. For testing the performance of ossPTMetaD, five protein systems with diverse biological functions and motion patterns were selected, including large-scale domain motion (AdK), flap movement (HIV-1 protease and BACE1), and DFG-motif flip in kinases (p38α and c-Abl). The simulation results showed that ossPTMetaD requires much fewer numbers of replicas than temperature REMD (T-REMD) with a reduction of ∼70% to achieve a similar exchange ratio. Although it does not obey the detailed balance condition, ossPTMetaD provides consistent results with T-REMD and experimental data. The high accessibility of the large conformational change of protein systems by ossPTMetaD, especially in simulating the very challenging DFG-motif flip of protein kinases, demonstrated its high efficiency and robustness in the characterization of the large-scale protein conformational change pathway and associated free energy profile.

Non-convex Learning via Replica Exchange Stochastic Gradient MCMC

Replica exchange Monte Carlo (reMC), also known as parallel tempering, is an important technique for accelerating the convergence of the conventional Markov Chain Monte Carlo (MCMC) algorithms. However, such a method requires the evaluation of the energy function based on the full dataset and is not scalable to big data. The naïve implementation of reMC in mini-batch settings introduces large biases, which cannot be directly extended to the stochastic gradient MCMC (SGMCMC), the standard sampling method for simulating from deep neural networks (DNNs). In this paper, we propose an adaptive replica exchange SGMCMC (reSGMCMC) to automatically correct the bias and study the corresponding properties. The analysis implies an acceleration-accuracy trade-off in the numerical discretization of a Markov jump process in a stochastic environment. Empirically, we test the algorithm through extensive experiments on various setups and obtain the state-of-the-art results on CIFAR10, CIFAR100, and SVHN in both supervised learning and semi-supervised learning tasks.

An Adaptive Empirical Bayesian Method for Sparse Deep Learning

We propose a novel adaptive empirical Bayesian (AEB) method for sparse deep learning, where the sparsity is ensured via a class of self-adaptive spike-and-slab priors. The proposed method works by alternatively sampling from an adaptive hierarchical posterior distribution using stochastic gradient Markov Chain Monte Carlo (MCMC) and smoothly optimizing the hyperparameters using stochastic approximation (SA). We further prove the convergence of the proposed method to the asymptotically correct distribution under mild conditions. Empirical applications of the proposed method lead to the state-of-the-art performance on MNIST and Fashion MNIST with shallow convolutional neural networks (CNN) and the state-of-the-art compression performance on CIFAR10 with Residual Networks. The proposed method also improves resistance to adversarial attacks.

Weak Convergence Rates of Population Versus Single-Chain Stochastic Approximation MCMC Algorithms

In this paper we establish the theory of weak convergence (toward a normal distribution) for both single-chain and population stochastic approximation Markov chain Monte Carlo (MCMC) algorithms (SAMCMC algorithms). Based on the theory, we give an explicit ratio of convergence rates for the population SAMCMC algorithm and the single-chain SAMCMC algorithm. Our results provide a theoretic guarantee that the population SAMCMC algorithms are asymptotically more efficient than the single-chain SAMCMC algorithms when the gain factor sequence decreases slower than O(1 / t), where t indexes the number of iterations. This is of interest for practical applications.

Weak Convergence Rates of Population Versus Single-Chain Stochastic Approximation MCMC Algorithms

In this paper we establish the theory of weak convergence (toward a normal distribution) for both single-chain and population stochastic approximation Markov chain Monte Carlo (MCMC) algorithms (SAMCMC algorithms). Based on the theory, we give an explicit ratio of convergence rates for the population SAMCMC algorithm and the single-chain SAMCMC algorithm. Our results provide a theoretic guarantee that the population SAMCMC algorithms are asymptotically more efficient than the single-chain SAMCMC algorithms when the gain factor sequence decreases slower than O(1 / t), where t indexes the number of iterations. This is of interest for practical applications.

Simulated tempering based on global balance or detailed balance conditions: Suwa-Todo, heat bath, and Metropolis algorithms

Simulated tempering (ST) is a useful method to enhance sampling of molecular simulations. When ST is used, the Metropolis algorithm, which satisfies the detailed balance condition, is usually applied to calculate the transition probability. Recently, an alternative method that satisfies the global balance condition instead of the detailed balance condition has been proposed by Suwa and Todo. In this study, ST method with the Suwa-Todo algorithm is proposed. Molecular dynamics simulations with ST are performed with three algorithms (the Metropolis, heat bath, and Suwa-Todo algorithms) to calculate the transition probability. Among the three algorithms, the Suwa-Todo algorithm yields the highest acceptance ratio and the shortest autocorrelation time. These suggest that sampling by a ST simulation with the Suwa-Todo algorithm is most efficient. In addition, because the acceptance ratio of the Suwa-Todo algorithm is higher than that of the Metropolis algorithm, the number of temperature states can be reduced by 25% for the Suwa-Todo algorithm when compared with the Metropolis algorithm.

A Multiresolution Method for Parameter Estimation of Diffusion Processes

Diffusion process models are widely used in science, engineering and finance. Most diffusion processes are described by stochastic differential equations in continuous time. In practice, however, data is typically only observed at discrete time points. Except for a few very special cases, no analytic form exists for the likelihood of such discretely observed data. For this reason, parametric inference is often achieved by using discrete-time approximations, with accuracy controlled through the introduction of missing data. We present a new multiresolution Bayesian framework to address the inference difficulty. The methodology relies on the use of multiple approximations and extrapolation, and is significantly faster and more accurate than known strategies based on Gibbs sampling. We apply the multiresolution approach to three data-driven inference problems - one in biophysics and two in finance - one of which features a multivariate diffusion model with an entirely unobserved component.

The molecular genetics of autism spectrum disorders: genomic mechanisms, neuroimmunopathology, and clinical implications

Autism spectrum disorders (ASDs) have become increasingly common in recent years. The discovery of single-nucleotide polymorphisms and accompanying copy number variations within the genome has increased our understanding of the architecture of the disease. These genetic and genomic alterations coupled with epigenetic phenomena have pointed to a neuroimmunopathological mechanism for ASD. Model animal studies, developmental biology, and affective neuroscience laid a foundation for dissecting the neural pathways impacted by these disease-generating mechanisms. The goal of current autism research is directed toward a systems biological approach to find the most basic genetic and environmental causes to this severe developmental disease. It is hoped that future genomic and neuroimmunological research will be directed toward finding the road toward prevention, treatment, and cure of ASD.

TIGER2: an improved algorithm for temperature intervals with global exchange of replicas

An empirical sampling method for molecular simulation based on "temperature intervals with global exchange of replicas" (TIGER2) has been developed to reduce the high demand for computational resources and the low computational efficiency of the conventional replica-exchange molecular dynamics (REMD) method. This new method overcomes the limitation of its previous version, called TIGER, which requires the assumption of constant heat capacity during quenching of replicas from elevated temperatures to the baseline temperature. The robustness of the TIGER2 method is examined by comparing it against a Metropolis Monte Carlo simulation for sampling the conformational distribution of a single butane molecule in vacuum, a REMD simulation for sampling the behavior of alanine dipeptide in explicit solvent, and REMD simulations for sampling the folding behavior of two peptides, (AAQAA)(3) and chignolin, in implicit solvent. The agreement between the results from these conventional sampling methods and the TIGER2 simulations indicates that the TIGER2 algorithm is able to closely approximate a Boltzmann-weighted ensemble of states for these systems but without the limiting assumptions that were required for the original TIGER algorithm. TIGER2 is an efficient replica-exchange sampling method that enables the number of replicas that are used for a replica-exchange simulation to be substantially reduced compared to the conventional REMD method.

Isothermal-isobaric molecular dynamics using stochastic velocity rescaling

The authors present a new molecular dynamics algorithm for sampling the isothermal-isobaric ensemble. In this approach the velocities of all particles and volume degrees of freedom are rescaled by a properly chosen random factor. The technical aspects concerning the derivation of the integration scheme and the conservation laws are discussed in detail. The efficiency of the barostat is examined in Lennard-Jones solid and liquid near the triple point and compared to the deterministic Nose-Hoover and the stochastic Langevin methods. In particular, the dependence of the sampling efficiency on the choice of the thermostat and barostat relaxation times is systematically analyzed.

Phylogenetic tree construction using sequential stochastic approximation Monte Carlo

Monte Carlo methods have received much attention recently in the literature of phylogenetic tree construction. However, they often suffer from two difficulties, the curse of dimensionality and the local-trap problem. The former one is due to that the number of possible phylogenetic trees increases at a super-exponential rate as the number of taxa increases. The latter one is due to that the phylogenetic tree has often a rugged energy landscape. In this paper, we propose a new phylogenetic tree construction method, which attempts to alleviate these two difficulties simultaneously by making use of the sequential structure of phylogenetic trees in conjunction with stochastic approximation Monte Carlo (SAMC) simulations. The use of the sequential structure of the problem provides substantial help to reduce the curse of dimensionality in simulations, and SAMC effectively prevents the system from getting trapped in local energy minima. The new method is compared with a variety of existing Bayesian and non-Bayesian methods on simulated and real datasets. Numerical results are in favor of the new method in terms of quality of the resulting phylogenetic trees.

Accurate sampling using Langevin dynamics

We show how to derive a simple integrator for the Langevin equation and illustrate how it is possible to check the accuracy of the obtained distribution on the fly, using the concept of effective energy introduced in a recent paper [J. Chem. Phys. 126, 014101 (2007)]. Our integrator leads to correct sampling also in the difficult high-friction limit. We also show how these ideas can be applied in practical simulations, using a Lennard-Jones crystal as a paradigmatic case.

Canonical sampling through velocity rescaling

The authors present a new molecular dynamics algorithm for sampling the canonical distribution. In this approach the velocities of all the particles are rescaled by a properly chosen random factor. The algorithm is formally justified and it is shown that, in spite of its stochastic nature, a quantity can still be defined that remains constant during the evolution. In numerical applications this quantity can be used to measure the accuracy of the sampling. The authors illustrate the properties of this new method on Lennard-Jones and TIP4P water models in the solid and liquid phases. Its performance is excellent and largely independent of the thermostat parameter also with regard to the dynamic properties.

Use of sequential structure in simulation from high-dimensional systems

Sampling from high-dimensional systems often suffers from the curse of dimensionality. In this paper, we explored the use of sequential structures in sampling from high-dimensional systems with an aim at eliminating the curse of dimensionality, and proposed an algorithm, so-called sequential parallel tempering as an extension of parallel tempering. The algorithm was tested with the witch's hat distribution and Ising model. Numerical results suggest that it is a promising tool for sampling from high-dimensional systems. The efficiency of the algorithm was argued theoretically based on the Rao-Blackwellization theorem.