Data efficient learning of molecular slow modes from nonequilibrium metadynamics DOI

J. Hanni,

Dhiman Ray

The Journal of Chemical Physics, Journal Year: 2025, Volume and Issue: 162(16)

Published: April 22, 2025

Enhanced sampling simulations help overcome free energy barriers and explore molecular conformational space by applying external bias potential along suitable collective variables (CVs). However, identifying optimal CVs that align with the slow modes of complex systems many coupled degrees freedom can be a significant challenge. Deep time-lagged independent component analysis (Deep-TICA) addresses this issue employing an artificial neural network generates non-linear combinations descriptors to learn slowest freedom. Training Deep-TICA CVs, however, typically requires long equilibrium sample multiple recrossing events across various metastable conformations molecule. This requirement often prohibitively expensive, thereby limiting its widespread application. In study, we present algorithm enables training using limited amount trajectory data obtained from short, non-equilibrium metadynamics only one forward transition initial final state. We achieve utilizing variational Koopman algorithm, which reweights short off-equilibrium trajectories reflect probability densities. demonstrate enhanced conducted reweighted CV accurately efficiently converge surface for such as Müller–Brown potential, alanine dipeptide, chignolin mini-protein. Our approach, therefore, key challenge inferring data, making it more feasible use deep learning study processes practical relevance.

Language: Английский

Computing the committor with the committor to study the transition state ensemble DOI
Pei‐Lin Kang, Enrico Trizio, Michele Parrinello

et al.

Nature Computational Science, Journal Year: 2024, Volume and Issue: 4(6), P. 451 - 460

Published: June 5, 2024

Language: Английский

Citations

21

How Does Structural Disorder Impact Heterogeneous Catalysts? The Case of Ammonia Decomposition on Non-stoichiometric Lithium Imide DOI
Francesco Mambretti, Umberto Raucci, Manyi Yang

et al.

ACS Catalysis, Journal Year: 2024, Volume and Issue: 14(3), P. 1252 - 1256

Published: Jan. 10, 2024

Among the many catalysts suggested for ammonia decomposition, Li2NH has been shown to be quite promising. In recent past, we have performed extensive ab initio-quality simulations explain workings of this unusual catalyst. complex scenario that emerged, surface dynamics and structural disorder enhanced by interaction with reacting molecules played crucial roles. Non-stoichiometric lithium imide (Li2–x(NH2)x(NH)1–x) reported better catalytic performances than pure imide. Stimulated these findings, follow up our previous study simulating decomposition on such non-stoichiometric compounds. We attribute reactivity fact compositional further enhances fluctuations in topmost layers catalyst, strengthening dynamic picture process.

Language: Английский

Citations

13

Collective Variable-Based Enhanced Sampling: From Human Learning to Machine Learning DOI
Haohao Fu, Hengwei Bian, Xueguang Shao

et al.

The Journal of Physical Chemistry Letters, Journal Year: 2024, Volume and Issue: 15(6), P. 1774 - 1783

Published: Feb. 8, 2024

Enhanced-sampling algorithms relying on collective variables (CVs) are extensively employed to study complex (bio)chemical processes that not amenable brute-force molecular simulations. The selection of appropriate CVs characterizing the slow movement modes is paramount importance for reliable and efficient enhanced-sampling In this Perspective, we first review application limitations obtained from chemical geometrical intuition. We also introduce path-sampling algorithms, which can identify path-like in a high-dimensional free-energy space. Machine-learning offer viable approach finding suitable by analyzing trajectories preliminary discuss both performance machine-learning-derived simulations experimental models challenges involved applying these realistic, assemblies. Moreover, provide prospective view potential advancements machine-learning development field

Language: Английский

Citations

10

PLUMED Tutorials: A collaborative, community-driven learning ecosystem DOI
Gareth A. Tribello, Massimiliano Bonomi, Giovanni Bussi

et al.

The Journal of Chemical Physics, Journal Year: 2025, Volume and Issue: 162(9)

Published: March 4, 2025

In computational physics, chemistry, and biology, the implementation of new techniques in shared open-source software lowers barriers to entry promotes rapid scientific progress. However, effectively training users presents several challenges. Common methods like direct knowledge transfer in-person workshops are limited reach comprehensiveness. Furthermore, while COVID-19 pandemic highlighted benefits online training, traditional tutorials can quickly become outdated may not cover all software’s functionalities. To address these issues, here we introduce “PLUMED Tutorials,” a collaborative model for developing, sharing, updating tutorials. This initiative utilizes repository management continuous integration ensure compatibility with updates. Moreover, interconnected form structured learning path enriched automatic annotations provide broader context. paper illustrates development, features, advantages PLUMED Tutorials, aiming foster an open community creating sharing educational resources.

Language: Английский

Citations

2

Deep learning path-like collective variable for enhanced sampling molecular dynamics DOI Creative Commons
Thorben Fröhlking, Luigi Bonati, Valerio Rizzi

et al.

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 160(17)

Published: May 2, 2024

Several enhanced sampling techniques rely on the definition of collective variables to effectively explore free energy landscapes. The existing that describe progression along a reactive pathway offer an elegant solution but face number limitations. In this paper, we address these challenges by introducing new path-like variable called “deep-locally non-linear-embedding,” which is inspired principles locally linear embedding technique and trained trajectory. mimics ideal reaction coordinate automatically generating non-linear combination features through differentiable generalized autoencoder combines neural network with continuous k-nearest neighbor selection. Among key advantages method its capability choose metric for searching neighbors learn path from state A B without need handpick landmarks priori. We demonstrate effectiveness DeepLNE showing closely approximates in toy models, such as Müller-Brown potential alanine dipeptide. Then, use it molecular dynamics simulations RNA tetraloop, where highlight accelerate transitions estimate folding.

Language: Английский

Citations

9

Effective data-driven collective variables for free energy calculations from metadynamics of paths DOI Creative Commons
Lukas Müllender, Andrea Rizzi, Michele Parrinello

et al.

PNAS Nexus, Journal Year: 2024, Volume and Issue: 3(4)

Published: March 28, 2024

A variety of enhanced sampling (ES) methods predict multidimensional free energy landscapes associated with biological and other molecular processes as a function few selected collective variables (CVs). The accuracy these is crucially dependent on the ability chosen CVs to capture relevant slow degrees freedom system. For complex processes, finding such real challenge. Machine learning (ML) offer, in principle, solution handle this problem. However, rely availability high-quality datasets-ideally incorporating information about physical pathways transition states-which are difficult access, therefore greatly limiting their domain application. Here, we demonstrate how datasets can be generated by means ES simulations trajectory space via metadynamics paths algorithm. approach expected provide general efficient way generate ML-based for fast prediction simulations. We our two numerical examples, 2D model potential isomerization alanine dipeptide, using deep targeted discriminant analysis CV choice.

Language: Английский

Citations

7

Combining Transition Path Sampling with Data-Driven Collective Variables through a Reactivity-Biased Shooting Algorithm DOI
Jintu Zhang,

Odin Zhang,

Luigi Bonati

et al.

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(11), P. 4523 - 4532

Published: May 27, 2024

Rare event sampling is a central problem in modern computational chemistry research. Among the existing methods, transition path (TPS) can generate unbiased representations of reaction processes. However, its efficiency depends on ability to reactive trial paths, which turn quality shooting algorithm used. We propose new based success rate, i.e., reactivity, measured as function reduced set collective variables (CVs). These are extracted with machine learning approach directly from TPS simulations, using multitask objective function. Iteratively, this workflow significantly improves without any prior knowledge process. In addition, optimized CVs be used biased enhanced methodologies accurately reconstruct free energy profiles. tested method three different systems: two-dimensional toy model, conformational transitions alanine dipeptide, and hydrolysis acetyl chloride bulk water. latter, we integrated our an active scheme learn learning-based potential, allowed us study mechanism profile ab initio-like accuracy.

Language: Английский

Citations

7

Acceleration with Interpretability: A Surrogate Model-Based Collective Variable for Enhanced Sampling DOI

Sompriya Chatterjee,

Dhiman Ray

Journal of Chemical Theory and Computation, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 4, 2025

Most enhanced sampling methods facilitate the exploration of molecular free energy landscapes by applying a bias potential along reduced dimensional collective variable (CV) space. The success these depends on ability CVs to follow relevant slow modes system. Intuitive CVs, such as distances or contacts, often prove inadequate, particularly in biological systems involving many coupled degrees freedom. Machine learning algorithms, especially neural networks (NN), can automate process CV discovery combining large number descriptors and outperform intuitive efficiency. However, their lack interpretability high cost evaluation during trajectory propagation make NN-CVs difficult apply biomolecular processes. Here, we introduce surrogate model approach using lasso regression express output network linear combination an automatically chosen subset input descriptors. We demonstrate successful applications our simulation conformational landscape alanine dipeptide chignolin mini-protein. In addition providing mechanistic insights due explainable nature, showed negligible loss efficiency accuracy, compared NN-CVs, reconstructing underlying surface. Moreover, simplified functional forms, are better at extrapolating unseen regions space, e.g., saddle points. Surrogate also less expensive evaluate NN counterparts, making them suitable for complex

Language: Английский

Citations

1

Everything everywhere all at once: a probability-based enhanced sampling approach to rare events DOI
Enrico Trizio, Pei‐Lin Kang, Michele Parrinello

et al.

Nature Computational Science, Journal Year: 2025, Volume and Issue: unknown

Published: May 5, 2025

Language: Английский

Citations

1

Biomolecular dynamics in the 21st century DOI Creative Commons
Charles L. Brooks, Alexander D. MacKerell, Carol Beth Post

et al.

Biochimica et Biophysica Acta (BBA) - General Subjects, Journal Year: 2023, Volume and Issue: 1868(2), P. 130534 - 130534

Published: Dec. 6, 2023

The relevance of motions in biological macromolecules has been clear since the early structural analyses proteins by X-ray crystallography. Computer simulations have applied to provide a deeper understanding dynamics 1976, and are now standard tool many labs working on structure function biomolecules. In this mini-review we highlight some areas current interest active development for simulations, particular all-atom molecular simulations.

Language: Английский

Citations

16