CARD: Coarse-to-fine Autoregressive Modeling with Radix-based Decomposition for Transferable Free Energy Estimation

2026-05-04 · Source: Machine Learning · Field: Science & Research — Artificial Intelligence & Machine Learning, Life Sciences & Biology, Physical Sciences & Chemistry · Depth: Expert, quick

Summary

CARD is a novel generative framework designed to estimate free energy differences in molecular interactions, a critical task in chemistry and drug discovery. It addresses limitations of existing methods, which are either computationally expensive (classical approaches) or lack generalization (deep learning methods tied to specific input dimensions). CARD employs a radix-based decomposition to convert 3D coordinates into mixed discrete-continuous sequences, facilitating coarse-to-fine autoregressive modeling with improved expressiveness. This model represents a distribution with zero free energy, functioning as a proposal for absolute free energy computation across arbitrary systems without needing alchemical pathways. Experiments show CARD achieves accuracy comparable to classical computational methods on unseen systems with varied topologies, while delivering an approximate 40-fold speedup in inference.

Key takeaway

For AI Scientists and Research Scientists developing computational chemistry tools, CARD presents a significant advancement. Its ability to accurately estimate free energy differences for arbitrary systems without alchemical pathways, coupled with a 40-fold inference speedup, means you can accelerate drug discovery and materials science research. Consider integrating CARD's radix-based decomposition and autoregressive modeling principles into your next-generation simulation platforms.

Key insights

CARD offers a generative framework for free energy estimation, combining radix-based decomposition with autoregressive modeling for enhanced generalization.

Principles

• Radix-based decomposition enables 3D coordinate conversion.
• Zero free energy distribution serves as an absolute free energy proposal.

Method

CARD converts 3D coordinates into mixed discrete-continuous sequences using radix-based decomposition, then applies coarse-to-fine autoregressive modeling to estimate free energy differences.

In practice

• Estimate free energy for unseen molecular systems.
• Accelerate molecular interaction analysis by 40x.

Topics

• CARD
• Free Energy Estimation
• Radix-based Decomposition
• Autoregressive Modeling
• Molecular Dynamics

Best for: AI Scientist, Research Scientist

Related on AIssential

• Force-free molecular dynamics through autoregressive equivariant networks — TrajCast uses equivariant MPNNs to directly predict molecular states, enabling faster, data-efficient MD simulations with strong generalization.
• AutoDFT: A Closed-Loop Multi-Agent Framework for Autonomous DFT Calculations — AutoDFT automates complex DFT calculations via a closed-loop multi-agent LLM framework, adapting dynamically to emergent results.
• Accelerating AI-Powered Chemistry and Materials Science Simulations with NVIDIA ALCHEMI Toolkit-Ops — NVIDIA ALCHEMI Toolkit-Ops accelerates atomistic simulations by providing GPU-optimized, batched operations for chemistry and materials science.
• A unified predictive and generative solution for liquid electrolyte formulation — A unified AI framework accelerates liquid electrolyte design by combining predictive and generative models.
• De novo molecular generation with optical property preconditioning at the token level — Token-conditioned autoregressive models can generate OLED molecules with targeted optical properties, but control varies by chemical motif.
• A Systematic Evaluation of Molecular Mixture Behavior Prediction — Evaluating molecular mixture ML models requires decomposing error into pure-compound and non-ideal interaction components, moving beyond absolute accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.