Biology subjects

Tiwary, P.

Publications and source records attributed to Tiwary, P..

At least 19 recordsLinked to original sources

bioRxiv2026

Phylogenetic evidence for the emergence of bacterial σ-factors by domain fusion

{sigma}-factors are dissociable subunits of RNA polymerase (RNAP), forming the basal regulatory machinery for bacterial transcription. The minimal {sigma}-factors contain two domains (called {sigma}2 or {sigma}4), both of which bind RNAP as well as the promoter DNA through their helix-turn-helix motif. Housekeeping {sigma}-factors contain four domains. We identify proteins containing at least one of the four {sigma}-factor domains in bacterial and bacteriophage genomes; we also show that proteins containing either {sigma}2 or {sigma}4, which are not defined as {sigma}-factors, are commonly found in both sets of organisms. Using phylogenetic analysis of these two domains, we suggest that the highly conserved major housekeeping {sigma}-factor, with its complex domain architecture, represents a late-evolving monophyletic clade, whereas single-domain proteins and minimal {sigma}-factor architectures comprising {sigma}2 and {sigma}4 might be ancestral. Thus, {sigma}-factors probably evolved from single-domain DNA-binding transcription factors by domain accretion.

evolutionary biology↗

bioRxiv2026

Kryptix-1: Conformational Ensemble Sampling Substantially Improves Cryptic Pocket Detection Relative to Static-Structure and Prior Computational Approaches

Cryptic binding pockets; sites absent or occluded in a proteins resting-state structure that become druggable only in specific, transiently populated conformations; represent one of the largest untapped opportunities in structure-based drug discovery. Large-scale structural surveys estimate that cryptic pockets occur in on the order of one in six protein families genome-wide, and that accounting for them could expand the druggable fraction of the disease-associated human proteome from roughly 40% to nearly 80%. Detecting cryptic pockets computationally has historically forced a choice btw physics-based conformational sampling, which is reliable but too computationally expensive to deploy across more than a handful of targets, and fast machine-learning pocket predictors trained on static structures, which frequently over-predict and lose the precision needed for practical triage. We report Kryptix-1, an in-house pipeline developed at Covenant Biosciences that combines conformational ensemble generation with a consensus pocket-scoring algorithm, benchmarked here against a matched static-structure baseline and against the published literature on a 10-protein sample from CryptoBench, an independently curated cryptic-pocket reference dataset. On the fields own standard residue-overlap metric (Jaccard index 0.5), Kryptix-1 succeeds on 80% of benchmark proteins, compared to 20% for the static baseline and approximately 40-45% for the best-performing methods reported in an independent comparative evaluation using the same metric. We report these results alongside an explicit accounting of where Kryptix-1s advantage is smaller or reverses, and we are explicit throughout that this is a small pilot evaluation, not a fully powered validation study.

bioinformatics↗

bioRxiv2026

Topology, Energy Landscapes and Folding Kinetics of RNA Hairpins

Many RNAs occupy multiple inter-converting structures in order to perform key biological functions. Characterizing RNA conformational ensembles is therefore critical to illuminating the mechanisms by which RNAs fold, unfold, and undergo precise structural rearrangements in response to cellular signals. However, resolving the equilibrium populations and folding kinetics of RNA ensembles is nontrivial due to the intrinsic ruggedness of RNA conformational landscapes and wide range of timescales spanned by RNA structural dynamics. To address these challenges, we present a de novo hierarchical multiscale computational framework that integrates diverse secondary and tertiary structure generation, adaptive unbiased molecular dynamics, a generative AI model termed latent thermodynamic flows (LaTF), and Markov state modeling to efficiently map global RNA folding landscapes and associated dynamics. Applied to the GCAA tetraloop, HIV-TAR stem-loop, and microROSE RNA thermometer, this framework yields ~ 1.3 ms of cumulative sampling and resolves temperature-dependent conformational landscapes comprising numerous native-like, misfolded, partially folded, and unfolded metastable states. Across all three RNA hairpins, temperature substantially reshapes landscape ruggedness, metastable state populations, and folding mechanisms. Our predicted ensembles show overall agreement with key static and dynamic biophysical observables, including Nuclear Magnetic Resonance (NMR)-resolved structures, Nuclear Overhauser Effect (NOE)-derived distance restraints, and residual dipolar couplings (RDCs). Together, these results establish a general physics-informed AI approach for accurately modeling RNA conformational heterogeneity, equilibrium thermodynamics, and long-timescale kinetics.

biophysics↗

bioRxiv2026

Mavchen 1: A Conformational Ensemble Platform for Protein Ligand Pose Prediction That Substantially Outperforms Static Structure Prediction in a Category-Stratified Benchmark

Deep learning structure predictors, most prominently AlphaFold2 (the field-standard tool benchmarked against throughout this study), have substantially expanded access to protein structural information, yet characteristically return a single static conformation per target. This is an incomplete representation of the binding-competent state for the many pharmacologically relevant targets whose recognition geometry is intrinsically dependent on receptor flexibility, including cryptic-pocket, induced-fit, and water-mediated binding mechanisms. We present a category-stratified, statistically powered benchmark comparing pose prediction from receptor conformational ensembles against AlphaFold2, used as a matched static-structure baseline, across 29 protein-ligand systems spanning cryptic-pocket, induced-fit, water-mediated, and autoimmune-indication target classes. Considering the most accurate pose available from each methods full candidate output, ensemble-derived poses achieved lower RMSD to the experimental structure than AlphaFold on 21 of 29 targets (72.4%), with a mean RMSD of 3.39 [A] versus 5.60 [A]: a clear, statistically decisive advantage (paired Wilcoxon signed-rank test, W = 93.0, p = 0.0060). Rather than being diffuse, this advantage was concentrated precisely where mechanistic theory predicts it should be: in induced-fit and water-mediated categories, the classes in which static-structure prediction is expected to be least representative of the bound state: a result that constitutes direct, quantitative confirmation of the ensemble hypothesis, not merely a favorable average. Independent assessment against a field-standard physical-validity framework confirmed that this accuracy gain was achieved without any trade-off in chemical or geometric realism. We further quantify, rather than assume, the extent to which this advantage is recoverable by fully autonomous pose selection, using a proprietary ensemble-aware scoring model with no access to the correct answer, and report a substantial, discriminative signal (cross-validated mean AUC 0.92) with a partial, and clearly characterized, recovery under the strictest accuracy criteria (mean AUPR 0.36), which we identify as the principal, now precisely quantified, determinant of near-term translational progress. Under this same fully autonomous, ground-truth-blind setting, AlphaFolds own top-ranked poses currently match or modestly exceed Mavchen-1s autonomously selected poses on strict success-rate criteria (e.g., 17.2% vs. 20.7% at the combined RMSD-and-validity threshold), a result we report without qualification as the clearest current benchmark for near-term development. Together, these results provide compelling, statistically rigorous evidence that conformational ensemble sampling is a mechanistically grounded and substantial source of improved pose accuracy relative to static-structure prediction, and establish a quantitative benchmark against which continued methodological development can be measured and demonstrably improved upon.

bioinformatics↗

bioRxiv2026

CHIMIYA-1: An Autoselection Foundation Model for ADMET Property Prediction, Rigorously Benchmarked Against the Therapeutics Data Commons ADMET Group

Accurate, generalizable prediction of absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties remains one of the highest-leverage unsolved problems in computational drug discovery, and late-stage attrition driven by ADMET liabilities continues to be a dominant cost driver in pharmaceutical research and development. The Therapeutics Data Commons (TDC) ADMET Group has emerged as the fields most widely adopted public benchmark, comprising 22 endpoints under standardized scaffold-split evaluation. In this work we report a comprehensive evaluation of CHIMIYA-1, a proprietary autoselection foundation model developed by Covenant Biosciences, against the full TDC ADMET Group. Departing from common practice in the field, every reported score is the mean and standard deviation of five independently seeded end-to-end evaluation runs (TDCs own minimum submission standard, which we find is not met by all public leaderboard entries), and all 22 endpoints were additionally subjected to an explicit train/test structural-overlap audit prior to reporting, finding zero overlaps on any endpoint. Despite this deliberately conservative evaluation standard, CHIMIYA-1 ranks first among all publicly listed methods on four endpoints, places within the top decile of the field on twenty of twenty-two endpoints (91%), and attains a mean percentile standing near the 74th percentile across the full benchmark, with particular strength on toxicity and physicochemical-property endpoints. We further show that several top-ranked public comparators on this benchmark have been independently found to exhibit confirmed data leakage, a finding that, if anything, understates CHIMIYA-1s relative standing. All results were obtained on commodity single-GPU workstation hardware without recourse to distributed or cloud-scale training infrastructure. We discuss these results in the context of benchmark reporting norms in molecular machine learning and outline ongoing extensions, including continuous prospective-data retraining and CUDA-level throughput optimization of the underlying selection pipeline.

bioinformatics↗

bioRxiv2026

MAHLER: Integrating Metadynamics and Inverse Folding to Predict Antibody-Antigen Kinetics

Binding kinetics are crucial for antibody function, shaping pharmacokinetics and in vivo efficacy beyond what equilibrium affinity captures. We present "Metadynamics-Anchored Hybrid Learning for Engineering off-Rates (MAHLER)", a fully open-source machine learning/physics hybrid method that predicts relative antibody-antigen residence times at scale. Incorporating inverse-folding models into molecular dynamics simulations, MAHLER shows first-in-class screening-grade accuracy in calculating relative antibody-antigen dissociation kinetics across a family of point mutants. After initial antigen-specific setup, each prediction takes only 4 minutes on a single NVIDIA A100 GPU, compared to days even with already enhanced molecular dynamics simulations. This provides practical kinetics-aware complement to current computational design approaches that focus primarily on binding affinity for antibody-antigen complexes.

biophysics↗

bioRxiv2026

Detecting and quantifying overparametrization in RNA language models with REDIAL

While RNA language models (LMs) have served as foundation models (FMs) to advanced structural prediction, their evaluation relies heavily on supervised downstream tasks. Such tasks can often mask FM inefficiencies and reflect downstream training set memorization. To address this, here we introduce REDIAL (RNA Embedding perturbation Diagnostics for Language models), a zero-shot, unsupervised framework designed to extract coevolutionary signals directly from the high-dimensional latent spaces of RNA language models. By applying REDIAL, we uncover stark, layer-wise disparities in how popular RNA language models (LMs) internalize structural constraints through a layer-wise dissection and ablation study. Our results showed how such layerwise behavior deviates from protein LMs and is related to design flaws in the architectures. Specifically, we show that current RNA LMs are severely overparameterized relative to the limited sequence diversity of available RNA databases, leading to profound parameter inefficiency and overfitting. Furthermore, we establish that structure-guided pretraining fundamentally improves the signal-to-noise ratio of learned coevolutionary couplings compared to sequence-only baselines. Ultimately, this unsupervised evaluation paradigm exposes critical flaws in current parameter scaling strategies and provides a rigorous diagnostic benchmark to guide the development of more efficient, generalizable foundation models for RNA therapeutics and de novo design.

bioinformatics↗

bioRxiv2026

Early-Enrichment Hit Discovery via Reversible-Work c(t) Estimation in Metadynamics (CTMD)

Virtual screening for small-molecule binders is often limited by false positives from approximate scoring functions and rigid-receptor assumptions. These can be addressed downstream through accurate but expensive free energy calculations. At the same time, recent artificial-intelligence-based co-folding methods have been proposed that claim to achieve accuracy of free energy methods at much lower cost, but these have not yet delivered consistent improvements in early enrichment and can be confounded by memorization. Here we address this gap by introducing c(t)-based metadynamics (CTMD), a physics-based, high-throughput hit-triaging protocol tailored for early enrichment. CTMD uses the nonequilibrium reversible-work estimator c(t) introduced by Tiwary and Parrinello (Journal of Physical Chemistry B, 2015 119 736), computed from a small number of short, independent well-tempered metadynamics trajectories, to rank binding stability without requiring converged binding free energies. Across diverse targets and chemotypes, CTMD provides robust early enrichment while remaining fast, transferable with minimal parameter tuning, and resistant to memorization-driven artifacts--underscoring both an immediately deployable physics-based alternative for screening. For these systems we show how co-folding, particularly Boltz-2, achieves enrichment directly proportional to similairty with training set, and more worrying, even in the presence of signficant modifications to active site. Given its simplicity of implementation, CTMD should thus be an "embarassingly" open-source, early enrichment method available for use by the broad pharma and academic community that sits right between approximate but fast docking or AI based co-folding methods, and more expensive but accurate free energy calculations, expected to lead to saving significant financial and human capital in drug discovery campaigns.

biophysics↗

bioRxiv2026

Ab initio prediction of RNA structure ensembles with RNAnneal

RNA utilizes three-dimensional structure in addition to sequence to carry out diverse functions in gene expression and disease. Much like well-folded proteins, RNAs adopt specific three-dimensional structures to carry out their function. Yet comparatively few RNA structures have been solved by atomic resolution structural techniques, in part because unlike structured proteins, RNAs fold into heterogeneous ensembles of interconverting structures that pose a challenge for high-resolution structure probing methods. In this work, we introduce RNAnneal as a method for RNA structural ensemble prediction that seamlessly integrates generative deep learning with statistical physics and molecular dynamics modeling. Given the primary sequence, RNAnneal uses ab inito (i.e., first principles) modeling to sample an ensemble of 3D structures, which, in turn, are used to train an ensemble of unsupervised deep learning models. The RNAnneal score, representing the consensus of the deep learning models, is then used to evaluate the 3D structures. We evaluated RNAnneal structures against 16 experimentally-resolved conformations (ERCs) of riboswitch RNAs and found that pseudoknot-free (PK-free) ERCs were well-reproduced by RNAnneal, with clear avenues for improving performance even on PK-comprising structures. Furthermore, we found that the RNAnneal score outperforms the Rosetta score and a state-of-the-art RNA forcefield on the task of classifying ERCs from decoys. We then introduce the interaction entropy as a measure of conformational heterogeneity within an ensemble and use it to assess our predictions. RNAnneal thus provides a generalizable framework for predicting RNA structural ensembles that will accelerate RNA-targeted drug discovery and the design of functional RNA molecules.

biophysics↗

bioRxiv2025

Using Time Dependent Rate Analysis to Evaluate the Quality of Machine Learned Reaction Coordinates for Biasing and Computing Kinetics

Having an accurate reaction coordinate (RC) is essential for reliable kinetic characterization of molecular processes, but there are few quantitative metrics to evaluate RC quality. In this study, we consider the dimensionless{gamma} metric from the Exponential Average Time-dependent Rate (EATR) method, which represents the fraction of a biasing potential along the RC that contributes to increasing the rate constant. We demonstrate that{gamma} can be used to test whether the utility of a reaction coordinate for predicting kinetics with a Meta-dynamics bias improves as the coordinate is iteratively updated to include new data. We evaluate reaction coordinates approximated via the iterative State Predictive Information Bottleneck (SPIB) approach, which was previously shown to be accurate across six protein-ligand dissociation systems. For these same systems, we compute{gamma} values and mean accelerated times [Formula]. After systematically scanning over fitting parameters, the results show that{gamma} increases closer to 1, while [Formula] decreases, revealing a consistent inverse correlation. These results demonstrate that{gamma} serves as a practical criterion for RC evaluation and offers guidance for selecting SPIB-derived coordinates yielding quantitative kinetic predictions.

biophysics↗

bioRxiv2024

A CRISPR homing screen finds a chloroquine resistance transporter-like protein of the Plasmodium oocyst essential for mosquito transmission of malaria

Genetic screens with barcoded PlasmoGEM vectors have identified thousands of Plasmodium gene functions in haploid blood stages, gametocytes and liver stages. However, the formation of diploid cells by fertilisation has hindered the use of genetic screens to investigate vector-parasite interactions during the mosquito stages of the parasite. In this study, we developed a scalable genetic system that uses barcoded gene targeting vectors equipped with a CRISPR-mediated homing mechanism to generate homozygous loss-of-function mutants to reveal gene functions in the functionally diploid life cycle stages. In this system, a knockout vector additionally expressing a gRNA for its target is integrated into one of the parental alleles and directs Cas9 to the intact allele after fertilisation, leading to its disruption. We find that this homing strategy is 90% effective in the oocyst, resulting in the generation of homozygous genotypes. A pilot screen reveals that PBANKA_0916000 encodes a chloroquine resistance transporter-like protein, CRTL, essential for oocyst growth and sporogony. The data point to an unexpected importance for the transmission of malaria of the poorly understood digestive vacuole of the oocyst that contains hemozoin crystals. The new screening strategy provides a method to discover systematically and at scale the essential malaria transmission genes whose first essential functions are after fertilisation in the bloodmeal, enabling their potential as targets for transmission-blocking interventions to be assessed.

microbiology↗

bioRxiv2024

Calculating Protein-Ligand Residence Times Through State Predictive Information Bottleneck based Enhanced Sampling

Understanding drug residence times in target proteins is key to improving drug efficacy and understanding target recognition in biochemistry. While drug residence time is just as important as binding affinity, atomiclevel understanding of drug residence times through molecular dynamics (MD) simulations has been difficult primarily due to the extremely long timescales. Recent advances in rare event sampling have allowed us to reach these timescales, yet predicting protein-ligand residence times remains a significant challenge. Here we present a semi-automated protocol to calculate the ligand residence times across 12 orders of magnitudes of timescales. In our proposed framework, we integrate a deep learning-based method, the state predictive information bottleneck (SPIB), to learn an approximate reaction coordinate (RC) and use it to guide the enhanced sampling method metadynamics. We demonstrate the performance of our algorithm by applying it to six different protein-ligand complexes with available benchmark residence times, including the dissociation of the widely studied anti-cancer drug Imatinib (Gleevec) from both wild-type Abl kinase and drug-resistant mutants. We show how our protocol can recover quantitatively accurate residence times, potentially opening avenues for deeper insights into drug development possibilities and ligand recognition mechanisms. TOC Graphic O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=107 SRC="FIGDIR/small/589710v1_ufig1.gif" ALT="Figure 1"> View larger version (27K): org.highwire.dtl.DTLVardef@11dc13borg.highwire.dtl.DTLVardef@79103dorg.highwire.dtl.DTLVardef@194b67org.highwire.dtl.DTLVardef@a570e3_HPS_FORMAT_FIGEXP M_FIG C_FIG

biophysics↗

bioRxiv2023

Graph Attention Site Prediction (GrASP): Identifying Druggable Binding Sites Using Graph Neural Networks with Attention

Identifying and discovering druggable protein binding sites is an important early step in computer-aided drug discovery but remains a difficult task where most campaigns rely on a priori knowledge of binding sites from experiments. Here we present a binding site prediction method called Graph Attention Site Prediction (GrASP) and re-evaluate assumptions in nearly every step in the site prediction workflow from dataset preparation to model evaluation. GrASP is able to achieve state-of-the-art performance at recovering binding sites in PDB structures while maintaining a high degree of precision which will minimize wasted computation in downstream tasks such as docking and free energy perturbation.

biophysics↗

bioRxiv2023

Learning high-dimensional reaction coordinates of fast-folding proteins using State Predictive Information Bottleneck and Bias Exchange Metadynamics

Biological events occurring on long timescales, such as protein folding, remain hard to capture with conventional molecular dynamics (MD) simulation. To overcome these limitations, enhanced sampling techniques can be used to sample regions of the free energy landscape separated by high energy barriers, thereby allowing to observe these rare events. However, many of these techniques require a priori knowledge of the appropriate reaction coordinates (RCs) that describe the process of interest. In recent years, Artificial Intelligence (AI) models have emerged as promising approaches to accelerate rare event sampling. However, integration of these AI methods with MD for automated learning of improved RCs is not trivial, particularly when working with undersampled trajectories and highly complex systems. In this study, we employed the State Predictive Information Bottleneck (SPIB) neural network, coupled with bias exchange metadynamics simulations (BE-metaD), to investigate the unfolding process of two proteins, chignolin and villin. By utilizing the high-dimensional RCs learned from SPIB even with poor training data, BE-metaD simulations dramatically accelerate the sampling of the unfolding process for both proteins. In addition, we compare different RCs and find that the careful selection of RCs is crucial to substantially speed up the sampling of rare events. Thus, this approach, leveraging the power of AI and enhanced sampling techniques, holds great promise for advancing our understanding of complex biological processes occurring on long timescales. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=139 SRC="FIGDIR/small/550401v1_ufig1.gif" ALT="Figure 1"> View larger version (29K): org.highwire.dtl.DTLVardef@18a9296org.highwire.dtl.DTLVardef@9dc2f3org.highwire.dtl.DTLVardef@16a0cf8org.highwire.dtl.DTLVardef@1799db1_HPS_FORMAT_FIGEXP M_FIG TABLE OF CONTENT GRAPHIC C_FIG

biophysics↗

bioRxiv2022

From sequence to Boltzmann weighted ensemble of structures with AlphaFold2-RAVE

While AlphaFold2 is rapidly being adopted as a new standard in protein structure predictions, it is limited to single structure prediction. This can be insufficient for the inherently dynamic world of biomolecules. Even with recent modifications towards conformational diversity, AlphaFold2 is devoid of providing thermodynamically ranked conformations. AlphaFold2-RAVE is an efficient protocol using the structural outputs from AlphaFold2 as initializations for AI augmented molecular dynamics. These simulations result in Boltzmann ranked ensembles, which we demonstrate on different proteins.

biophysics↗

bioRxiv2021

Interrogating RNA-small molecule interactions with structure probing and AI augmented-molecular simulations

While there is increasing interest in the study of RNA as a therapeutic target, efforts to understand RNA-ligand recognition at the molecular level lag far behind our understanding of protein-ligand recognition. This problem is complicated due to the more than ten orders of magnitude in timescales involved in RNA dynamics and ligand binding events, making it not straightforward to design experiments or simulations. Here we make use of artificial intelligence (AI)-augmented molecular dynamics simulations to directly observe ligand dissociation for cognate and synthetic ligands from a riboswitch system. The site-specific flexibility profiles from our simulations are compared with in vitro measurements of flexibility using Selective 2 Hydroxyl Acylation analyzed by Primer Extension and Mutational Profiling (SHAPE-MaP). Our simulations reproduce known relative binding affinities for the cognate and synthetic ligands, and pinpoint how both ligands make use of different aspects of riboswitch flexibility. On the basis of our dissociation trajectories, we also make and validate predictions of pairs of mutations for both the ligand systems that would show differing binding affinities. These mutations are distal to the binding site and could not have been predicted solely on the basis of structure. The methodology demonstrated here shows how molecular dynamics simulations with all-atom force-fields have now come of age in making predictions that complement existing experimental techniques and illuminate aspects of systems otherwise not trivial to understand.

biophysics↗

bioRxiv2021

Protein flexibility and dissociation pathway differentiation can explain onset of resistance mutations in kinases

Understanding how point mutations can render a ligand or a drug ineffective against a given biological target is a problem of immense fundamental and practical relevance. Often the efficacy of such resistance mutations can be explained purely on a thermo-dynamic basis wherein the mutated system displays a reduced binding affinity for the ligand. However, the more perplexing and harder to explain situation is when two protein sequences have the same binding affinity for a drug. In this work, we demonstrate how all-atom molecular dynamics simulations, specifically using recent developments grounded in statistical mechanics and information theory, can provide a detailed mechanistic rationale for such variances. We establish the dissociation mechanism for the popular anti-cancer drug Imatinib (Gleevec) against wild-type and N387S mutant of Abl kinase. We show how this single point mutation triggers a non-local response in the proteins flexibility and eventually leads to pathway differentiation during dissociation. This pathway differentiation explains why Gleevec has a long residence time in the wild-type Abl, but for the mutant, by opening up a backdoor pathway for ligand exit, an order of magnitude shorter residence time is obtained. We thus believe that this work marks an efficient and scalable approach to pinpoint the molecular determinants of resistance mutations in biomolecular receptors of pharmacological relevance that are hard to explain using a simple structural perspective and require mechanistic and kinetic insights. Significance statementRelapse in late-stage cancer patients is often correlated with the onset of drug resistance mutations. Some of these mutations are very far from the binding site and thus hard to explain from a purely structural perspective. Here we employ all-atom molecular dynamics simulations aided by ideas from information theory that can reach timescales of seconds with minimal human bias in how the sampling is enhanced. Through these we explain how a single point mutation triggers a non-local response in the protein kinases flexibility and eventually leads to pathway differentiation during dissociation, thereby significantly reducing the residence time of the drug.

biochemistry↗

bioRxiv2021

Making high-dimensional molecular distribution functions tractable through Belief Propagation on Factor Graphs

Molecular dynamics (MD) simulations provide a wealth of high-dimensional data at all-atom and femtosecond resolution but deciphering mechanistic information from this data is an ongoing challenge in physical chemistry and biophysics. Theoretically speaking, joint probabilities of the equilibrium distribution contain all thermodynamic information, but they prove increasingly difficult to compute and interpret as the dimensionality increases. Here, inspired by tools in probabilistic graphical modeling, we develop a factor graph trained through belief propagation that helps factorize the joint probability into an approximate tractable form that can be easily visualized and used. We validate the study through the analysis of the conformational dynamics of two small peptides with 5 and 9 residues. Our validations include testing the conditional dependency predictions through an intervention scheme inspired by Judea Pearl. Secondly we directly use the belief propagation based approximate probability distribution as a high-dimensional static bias for enhanced sampling, where we achieve spontaneous back-and-forth motion between metastable states that is up to 350 times faster than unbiased MD. We believe this work opens up useful ways to thinking about and dealing with high-dimensional molecular simulations.

biophysics↗