Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2020.06.11.146985

A statistical physics based approach to capture spurious solutions in AI-enhanced molecular dynamics

Pant, S.·Smith, Z.·Wang, Y.·Tajkhorshid, E.·Tiwary, P.
View original source ↗Export metadata ↓

Abstract

Artificial intelligence (AI)-based approaches have had indubitable impact across the sciences through the ability to extract relevant information from raw data. Recently AI has also seen use for enhancing the efficiency of molecular simulations, wherein AI derived slow modes are used to accelerate the simulation in targeted ways. However, while typical fields where AI is used are characterized by a plethora of data, molecular simulations per-construction suffer from limited sampling and thus limited data. As such the use of AI in molecular simulations can suffer from a dangerous situation where the AI-optimization could get stuck in spurious regimes, leading to incorrect characterization of the reaction coordinate (RC) for the problem at hand. When such an incorrect RC is then used to perform additional simulations, one could start to deviate progressively from the ground truth. To deal with this problem of spurious AI-solutions, here we report a novel and automated algorithm using ideas from statistical mechanics. It is based on the notion that a more reliable AI-solution will be one that maximizes the time-scale separation between slow and fast processes. To learn this time-scale separation even from limited data, we use a maximum caliber-based framework. We show the applicability of this automatic protocol for 3 classic benchmark problems, namely the conformational dynamics of a model peptide, ligand-unbinding from a protein, and folding/unfolding energy landscape of the C-terminal domain of protein G. We believe our work will lead to increased and robust use of trustworthy AI in molecular simulations of complex systems.

Source connections

Explore related subjects

biophysicsPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Pant, S., Smith, Z., Wang, Y., Tajkhorshid, E., Tiwary, P.. 2020-06-12. A statistical physics based approach to capture spurious solutions in AI-enhanced molecular dynamics. https://doi.org/10.1101/2020.06.11.146985

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Frequency-Dependent Structural Responses of Amyloid-β Fibrils to Terahertz Pulses: A Nonequilibrium Molecular Dynamics Study

The abnormal aggregation of amyloid proteins into fibrils is closely associated with several neurodegenerative diseases, highlighting the importance of developing effective strategies to modulate fibril structural stability. Here, we employed nonequilibrium molecular dynamics(NEMD) simulations to examine the effects of intense terahertz-to-mid-infrared pulses on three amyloid fibrils (2LMP, 2MXU, and 5OQV), focusing on the combined effects of pulse frequency and electric-field strength. Frequency scans from 0.5 to 60 THz in 0.5 THz steps revealed frequency-dependent losses of {beta}-sheet content, with several common response patterns in these fibrils. Comparisons at 2, 23, 33, and 50 THz showed that pulse-induced changes in secondary structure differed from those in high-temperature controls. Hydrogen-bond analysis showed that total hydrogen-bond counts could partially recover even as interchain hydrogen bonds continued to decrease, with distinct responses among representative chain pairs. We therefore propose a two-threshold electric-field framework for fibril disruption: Eth1 describes the onset of structural perturbation, whereas Eth2 indicates persistent structural loss after the pulse. This framework offers a testable approach for characterizing amyloid structural responses to external fields, although the underlying energy-coupling mechanisms, quantitative thresholds, and long-term recovery require further validation. These findings may provide a theoretical basis for exploring external-field-based strategies to modulate protein aggregation.

biophysics↗
bioRxiv2026

Organised streaming of cerebrospinal fluid in the human third ventricle

Quasi-periodic pressure pulses associated with cardiac and respiratory activity generate reciprocating cerebrospinal fluid (CSF) motion in the human ventricles. Regular waves dominated by the cardiac cycle are overlaid with cycles of low-frequency respiratory waves, the latter to become stronger in forced breathing or even dominated by large successive flushes in NREM sleep. The present study investigates how such waves contribute to residual streaming within the human ventricles. Transient CFD simulations were performed in the third ventricle at varying parameters of frequency, stroke volume and wave-form asymmetry. Increasing the aqueduct peak Reynolds-number, either by small stroke volume and high frequency or large stroke volume at low frequency, transfers the flow from a reversible flow pattern to the formation of clear preferential pathways, alternating between the in- and outflow cycle. The resulting permanent streaming flux across the aqueduct relative to the mean oscillatory flow-rate, i.e. the dimensionless streaming ratio SR follows a rational saturation relation, SR [~] ReAq,peak/(C + ReAq,peak), demonstrating that increasing the peak aqueduct Reynolds number yields diminishing returns and bounds the relative contribution of residual inter-ventricular streaming. These findings demonstrate differences in hydrodynamic organisation in the third ventricle and highlight the importance of this organisation for intraventricular fluid exchange.

biophysics↗
bioRxiv2026

Amyloid stability and condensate material properties independently govern the hnRNPA2 condensate to fibril transition

hnRNPA2 is a component of cellular membraneless organelles many of which contain RNA and RNA-binding proteins (RBPs). RBPs frequently contain low-complexity domains (LCDs) that can also assemble into amyloid fibrils. Fibrils isolated from patients with frontotemporal lobar degeneration contain ordered segments of these LCDs and disease-associated mutations can enhance the homotypic interactions that promote aggregation. Although previous studies have largely examined the condensate-to-fibril transition by altering condensate stability, how amyloid structure governs this process remains poorly understood. Here, we show that the hnRNPA2 LCD forms a reproducible fibril fold and by combining structure-guided thermodynamic profiling, peptide design, mutagenesis and in vitro characterization under conditions with (two-phase) and without (one-phase) condensates, we define the molecular determinants of its assembly. We identify an aromatic triad within the fibril core that controls fibril formation kinetics and thermodynamic stability. Mutational analysis further reveals that condensate stability and fibril stability are largely independent. Finally, microrheology linked reduced condensate viscosity to faster fibril formation, supporting a role for condensate material properties in regulating aggregation kinetics. Together, these findings support a model in which the condensate-to-fibril transition is governed by the interplay between a disordered ''fuzzy coat'' and a fibril-forming aromatic triad within hnRNPA2. This work provides a structural framework for understanding how condensates can transition into amyloid assemblies.

biophysics↗