How AI Reveals the Rapid Activation of Catch‑Bonds Under Mechanical Force

Background

Catch‑bonds are molecular interactions whose lifetime increases with applied force. Understanding the structural dynamics that enable this behavior has been challenging.

Methods

The study investigated the XDoc:CohE protein complex, a highly stable interface in cellulose‑degrading bacteria, using a combination of steered molecular dynamics (SMD), dynamic network analysis, and machine‑learning regression.

200 independent SMD simulations were performed, generating 2,200 trajectory fragments covering a distribution of rupture forces.

Dynamic network analysis identified neighboring residues and computed α‑carbon motion correlations.

Residue pairs with average correlation greater than 0.2 were selected as features for regression models.

Several regression algorithms were evaluated; support‑vector regression (SVR) achieved the lowest mean absolute percentage error (MAPE ≈ 10.6%).

Key Findings

Instantaneous activation: The catch‑bond in XDoc:CohE becomes active almost immediately after force is applied, rather than gradually.

Low global correlation: Across all 200 simulations, no single residue pair maintained an average correlation above 0.2, indicating that bond stability does not rely on a fixed set of contacts.

Predictive power of early‑stage data: Models trained on short‑window correlation data—up to 2.1 ns before rupture—accurately predicted rupture forces. SVR yielded a minimum MAPE of 10.6% (average 9.5 % ± 0.8 %).

Model robustness: All tested models reliably ranked mechanical stability across different force regimes. Prediction accuracy improved with increasing force despite a weak overall correlation (R ≈ 0.37) between total network correlation and rupture force.

Implications

The results demonstrate that artificial‑intelligence models can extract subtle dynamical signatures from brief simulation fragments, providing a new pathway to understand and predict force‑sensitive biomolecular interactions such as catch‑bonds. This dynamic‑focused approach may accelerate the design of biomimetic materials and inform therapeutic strategies targeting mechanotransduction.

References

Original paper: https://pubs.acs.org/doi/10.1021/acs.jctc.5c01181

Related news article: https://phys.org/news/2025-09-ai-uncovers-hidden-nature-toughest.html

Figures

Code example

来源：ScienceAI
本文
约1000字
，建议阅读
5
分钟
研究系统地揭示了 XDoc:CohE 界面在机械应力下的复杂动态，阐明了这种逆锁键在不同力度下的表现。