Learning Collective Variables from Time-lagged Generation

2025-07-09

AI Summary

Rare events in molecular dynamics—such as conformational transitions—are challenging to sample directly due to their long timescales. Existing machine learning approaches for collective variable (CV) learning primarily emphasize static discrimination between metastable states, failing to effectively encode slow dynamical evolution. To address this, we propose the Time-Lagged Conditioning (TLC) framework, the first to incorporate time-lagged conditional distributions into generative CV learning. TLC explicitly models system dynamics—not just static state separation—by enforcing temporal consistency in the learned CV space. Integrated with enhanced sampling methods—including steered molecular dynamics (SMD) and on-the-fly probability-enhanced sampling (OPES)—TLC is validated on the alanine dipeptide system. Results show that TLC-derived CVs significantly improve transition path characterization accuracy and sampling efficiency. In two distinct enhanced sampling tasks, TLC matches or surpasses state-of-the-art methods. This work establishes a new, interpretable, and dynamics-aware paradigm for rare-event modeling.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Multiagent Systems: Multiagent Learning
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Web Mining and Content Analysis: Machine learning and data science for the Web

Abstract

Rare events such as state transitions are difficult to observe directly with molecular dynamics simulations due to long timescales. Enhanced sampling techniques overcome this by introducing biases along carefully chosen low-dimensional features, known as collective variables (CVs), which capture the slow degrees of freedom. Machine learning approaches (MLCVs) have automated CV discovery, but existing methods typically focus on discriminating meta-stable states without fully encoding the detailed dynamics essential for accurate sampling. We propose TLC, a framework that learns CVs directly from time-lagged conditions of a generative model. Instead of modeling the static Boltzmann distribution, TLC models a time-lagged conditional distribution yielding CVs to capture the slow dynamic behavior. We validate TLC on the Alanine Dipeptide system using two CV-based enhanced sampling tasks: (i) steered molecular dynamics (SMD) and (ii) on-the-fly probability enhanced sampling (OPES), demonstrating equal or superior performance compared to existing MLCV methods in both transition path sampling and state discrimination.

Problem

• Learning collective variables for rare event observation
• Improving CV accuracy in dynamic behavior capture
• Validating CVs in enhanced sampling tasks

Innovation

• Learns CVs from time-lagged generative model
• Models time-lagged conditional distribution dynamics
• Validated on Alanine Dipeptide with SMD and OPES