Genetic Algorithm for Automated Parameterization of Network Hamiltonian Models of Amyloid Fibril Formation.

The time scales of long-time atomistic molecular dynamics simulations are typically reported in microseconds, while the time scales for experiments studying the kinetics of amyloid fibril formation are typically reported in minutes or hours. This time scale deficit of roughly 9 orders of magnitude presents a major challenge in the design of computer simulation methods for studying protein aggregation events. Coarse-grained molecular simulations offer a computationally tractable path forward for exploring the molecular mechanism driving the formation of these structures, which are implicated in diseases such as Alzheimer's, Parkinson's, and type-II diabetes. Network Hamiltonian models of aggregation are centered around a Hamiltonian function that returns the total energy of a system of aggregating proteins, given the graph structure of the system as an input. In the graph, or network, representation of the system, each protein molecule is represented as a node, and noncovalent bonds between proteins are represented as edges. The parameter, i.e., a set of coefficients that determine the degree to which each topological degree of freedom is favored or disfavored, must be determined for each network Hamiltonian model, and is a well-known technical challenge. The methodology is first demonstrated by beginning with an initial set of randomly parametrized models of low fibril fraction (<5% fibrillar), and evolving to subsequent generations of models, ultimately leading to high fibril fraction models (>70% fibrillar). The methodology is also demonstrated by applying it to optimizing previously published network Hamiltonian models for the 5 key amyloid fibril topologies that have been reported in the Protein Data Bank (PDB). The models generated by the AI produced fibril fractions that surpass previously published fibril fractions in 3 of 5 cases, including the most naturally abundant amyloid fibril topology, the 1,2 2-ribbon, which features a steric zipper. The authors also aim to encourage more widespread use of the network Hamiltonian methodology for fitting a wide variety of self-assembling systems by releasing a free open-source implementation of the genetic algorithm introduced here.

Neural Upscaling from Residue-Level Protein Structure Networks to Atomistic Structures.

Coarse-graining is a powerful tool for extending the reach of dynamic models of proteins and other biological macromolecules. Topological coarse-graining, in which biomolecules or sets thereof are represented via graph structures, is a particularly useful way of obtaining highly compressed representations of molecular structures, and simulations operating via such representations can achieve substantial computational savings. A drawback of coarse-graining, however, is the loss of atomistic detail-an effect that is especially acute for topological representations such as protein structure networks (PSNs). Here, we introduce an approach based on a combination of machine learning and physically-guided refinement for inferring atomic coordinates from PSNs. This "neural upscaling" procedure exploits the constraints implied by PSNs on possible configurations, as well as differences in the likelihood of observing different configurations with the same PSN. Using a 1 µs atomistic molecular dynamics trajectory of Aß1-40, we show that neural upscaling is able to effectively recapitulate detailed structural information for intrinsically disordered proteins, being particularly successful in recovering features such as transient secondary structure. These results suggest that scalable network-based models for protein structure and dynamics may be used in settings where atomistic detail is desired, with upscaling employed to impute atomic coordinates from PSNs.

Network Hamiltonian models reveal pathways to amyloid fibril formation.

Amyloid fibril formation is central to the etiology of a wide range of serious human diseases, such as Alzheimer's disease and prion diseases. Despite an ever growing collection of amyloid fibril structures found in the Protein Data Bank (PDB) and numerous clinical trials, therapeutic strategies remain elusive. One contributing factor to the lack of progress on this challenging problem is incomplete understanding of the mechanisms by which these locally ordered protein aggregates self-assemble in solution. Many current models of amyloid deposition diseases posit that the most toxic species are oligomers that form either along the pathway to forming fibrils or in competition with their formation, making it even more critical to understand the kinetics of fibrillization. A recently introduced topological model for aggregation based on network Hamiltonians is capable of recapitulating the entire process of amyloid fibril formation, beginning with thousands of free monomers and ending with kinetically accessible and thermodynamically stable amyloid fibril structures. The model can be parameterized to match the five topological classes encompassing all amyloid fibril structures so far discovered in the PDB. This paper introduces a set of network statistical and topological metrics for quantitative analysis and characterization of the fibrillization mechanisms predicted by the network Hamiltonian model. The results not only provide insight into different mechanisms leading to similar fibril structures, but also offer targets for future experimental exploration into the mechanisms by which fibrils form.

Comparative Exploratory Analysis of Intrinsically Disordered Protein Dynamics Using Machine Learning and Network Analytic Methods.

Simulations of intrinsically disordered proteins (IDPs) pose numerous challenges to comparative analysis, prominently including highly dynamic conformational states and a lack of well-defined secondary structure. Machine learning (ML) algorithms are especially effective at discriminating among high-dimensional inputs whose differences are extremely subtle, making them well suited to the study of IDPs. In this work, we apply various ML techniques, including support vector machines (SVM) and clustering, as well as related methods such as principal component analysis (PCA) and protein structure network (PSN) analysis, to the problem of uncovering differences between configurational data from molecular dynamics simulations of two variants of the same IDP. We examine molecular dynamics (MD) trajectories of wild-type amyloid beta (Aß1-40) and its "Arctic" variant (E22G), systems that play a central role in the etiology of Alzheimer's disease. Our analyses demonstrate ways in which ML and related approaches can be used to elucidate subtle differences between these proteins, including transient structure that is poorly captured by conventional metrics.

Predicting Reaction Products and Automating Reactive Trajectory Characterization in Molecular Simulations with Support Vector Machines.

A machine learning-based methodology for the prediction of chemical reaction products, along with automated elucidation of mechanistic details via phase space analysis of reactive trajectories, is introduced using low-dimensional heuristic models and then applied to ab initio computer simulations of the photodissociation of acetaldehyde, an important chemical system in atmospheric chemistry. Our method is centered around training Support Vector Machines (SVMs) to identify optimal separatrices that delineate the regions of phase space that lead to different chemical reaction products. In contrast to the more common "black box" type machine learning methodologies for analyzing chemical simulation data, this SVM-based methodology allows for mechanistic insight to be gleaned from further analysis of the positioning of the phase space points used to train the SVM with respect to the separatrices. For example, a pair of phase space points that are in close proximity to each other but on opposite sides of a separatrix may be situated on opposite sides of a transition state, while phase space points occurring early in a simulation that are distant from a separatrix are likely to belong to trajectories that are highly biased toward the product state associated with the basin of attraction to which they belong. In addition to inferring mechanistic details about multiple-pathway chemical reactions, our method can also be used to increase reactive trajectory sampling efficiency in molecular simulations via rejection sampling, with trajectories leading to undesired product states being identified and terminated early in the simulation rather than being carried to completion.

Network-Based Classification and Modeling of Amyloid Fibrils.

Amyloid fibrils are locally ordered protein aggregates that self-assemble under a variety of physiological and in vitro conditions. Their formation is of fundamental interest as a physical chemistry problem and plays a central role in Alzheimer's disease, Type II diabetes, and other human diseases. As the number of known amyloid fibril structures has grown, the need has arisen for a nomenclature for describing and classifying fibril types, as well as a theoretical description of the physics that gives rise to the self-assembly of these structures. Here, we introduce a systematic nomenclature and coarse-graining methodology for describing the topology of fibrils and other protein aggregates, along with a computational methodology for simulating protein aggregation. Both have mathematical underpinnings in graph theory and statistical mechanics and are consistent with available experimental data on the fibril structure and aggregation kinetics. Our graph representation of the fibril topology enables us to define a network Hamiltonian based on connectivity patterns among monomers rather than detailed intermolecular interactions, greatly speeding up the simulation of large ensembles. Our simulation strategy is capable of recapitulating the formation of all currently known amyloid fibril topologies found in the Protein Data Bank, as well as the formation kinetics of fibrils and oligomers.

Advances in milestoning. II. Calculating time-correlation functions from milestoning using stochastic path integrals.

In the milestoning framework, and more generally in related transition interface sampling schemes, one significantly enhances the calculation of relaxation rates for complex equilibrium kinetics from molecular dynamics simulations between the milestones or interfaces. The goal of the present paper is to advance milestoning applications into the realm of non-equilibrium statistical mechanics, in particular, to calculate entire time correlation functions. In order to accomplish this, we introduce a novel methodology for obtaining the flux through a given milestone configuration as a function of both time and initial configuration and build upon it with a novel formalism describing autocorrelation for Langevin motion in a discrete configuration space. The method is then applied to three different test systems: a harmonic oscillator, which we solve analytically, a two-well potential, which is solved numerically, and an atomistic molecular dynamics simulation of alanine dipeptide.

Advances in milestoning. I. Enhanced sampling via wind-assisted reweighted milestoning (WARM).

The milestoning algorithm of Elber and co-workers creates a framework for computing the time scale of processes that are too long and too complex to be studied using simply brute force simulations. The fundamental objects involved in the milestoning algorithm are the first passage time distributions KAB (τ) between adjacent conformational milestones A and B. The method proposed herein aims to further enhance milestoning (or other interface based sampling methods) by employing an artificially applied force, akin to a wind that blows the trajectories from their initial to their final states, and by subsequently applying corrective weights to the trajectories to yield the true first passage time distributions KAB (τ) in a fraction of the computation time required for unassisted calculations. The re-weighting method is rooted in the formalism of stochastic path integrals. The theoretical basis for the technique and numerical examples are presented.

Automated placement of interfaces in conformational kinetics calculations using machine learning.

Several recent implementations of algorithms for sampling reaction pathways employ a strategy for placing interfaces or milestones across the reaction coordinate manifold. Interfaces can be introduced such that the full feature space describing the dynamics of a macromolecule is divided into Voronoi (or other) cells, and the global kinetics of the molecular motions can be calculated from the set of fluxes through the interfaces between the cells. Although some methods of this type are exact for an arbitrary set of cells, in practice, the calculations will converge fastest when the interfaces are placed in regions where they can best capture transitions between configurations corresponding to local minima. The aim of this paper is to introduce a fully automated machine-learning algorithm for defining a set of cells for use in kinetic sampling methodologies based on subdividing the dynamical feature space; the algorithm requires no intuition about the system or input from the user and scales to high-dimensional systems.

m(1)A and m(1)G disrupt A-RNA structure through the intrinsic instability of Hoogsteen base pairs.

The B-DNA double helix can dynamically accommodate G-C and A-T base pairs in either Watson-Crick or Hoogsteen configurations. Here, we show that G-C(+) (in which + indicates protonation) and A-U Hoogsteen base pairs are strongly disfavored in A-RNA. As a result,N(1)-methyladenosine and N(1)-methylguanosine, which occur in DNA as a form of alkylation damage and in RNA as post-transcriptional modifications, have dramatically different consequences. Whereas they create G-C(+) and A-T Hoogsteen base pairs in duplex DNA, thereby maintaining the structural integrity of the double helix, they block base-pairing and induce local duplex melting in RNA. These observations provide a mechanism for disrupting RNA structure through post-transcriptional modifications. The different propensities to form Hoogsteen base pairs in B-DNA and A-RNA may help cells meet the opposing requirements of maintaining genome stability, on the one hand, and of dynamically modulating the structure of the epitranscriptome, on the other.

Rate turnover in mechano-catalytic coupling: A model and its microscopic origin.

A novel aspect in the area of mechano-chemistry concerns the effect of external forces on enzyme activity, i.e., the existence of mechano-catalytic coupling. Recent experiments on enzyme-catalyzed disulphide bond reduction in proteins under the effect of a force applied on the termini of the protein substrate reveal an unexpected biphasic force dependence for the bond cleavage rate. Here, using atomistic molecular dynamics simulations combined with Smoluchowski theory, we propose a model for this behavior. For a broad range of forces and systems, the model reproduces the experimentally observed rates by solving a reaction-diffusion equation for a "protein coordinate" diffusing in a force-dependent effective potential. The atomistic simulations are used to compute, from first principles, the parameters of the model via a quasiharmonic analysis. Additionally, the simulations are also used to provide details about the microscopic degrees of freedom that are important for the underlying mechano-catalysis.