Skip to content

ASTEROID: A Spatiotemporal Information Transformer for Forecasting Multi-Step Time Series of Molecular Dynamics

Home / Papers / ASTEROID: A Spatiotemporal Information Transformer for Forecasting Multi-Step Time Series of Molecular Dynamics

ASTEROID: A Spatiotemporal Information Transformer for Forecasting Multi-Step Time Series of Molecular Dynamics

Authors: Kexin Wu, Luonan Chen, Renxiao Wang Date: 2026-06-16 Paper ID: openalex:2606.17668

Summary

ASTEROID is a data-driven framework designed to accelerate molecular dynamics (MD) simulations by predicting multi-step atomic coordinates directly. By reformulating MD trajectories as high-dimensional spatiotemporal sequences and incorporating a novel Spatiotemporal Information (STI) Transformation into a Transformer architecture, the model efficiently handles complex multiscale dependencies. It employs a local-global self-attention mechanism for spatial interaction modeling and an encoder-decoder structure for robust temporal forecasting. Evaluation on quantum-mechanics derived molecular datasets confirms that the approach improves prediction accuracy while substantially reducing the computational overhead of traditional simulation methods.

Key Contributions

  • Introduces ASTEROID, a framework that models molecular dynamics trajectories as high-dimensional spatiotemporal sequences, enabling direct multi-step atomic coordinate prediction without iterative integration.
  • Implements a local-global self-attention mechanism to capture both short- and long-range spatial interactions in molecular systems.
  • Demonstrates superior prediction accuracy and significantly lower computational cost compared to conventional molecular dynamics integration methods on quantum-mechanics derived benchmarks.

Key Concepts

Archivist Review

I approved the STI Transformation as a concept because it represents a specific, reusable architectural design pattern for embedding domain-specific evolution rules into attention-based models. The ASTEROID framework itself was rejected as a local branding choice for an MD application rather than a generalized contribution. No datasets or open questions met the high bar for permanent archival given the paper’s focus on a specific domain application.

Approved Concepts

  • Spatiotemporal Information (STI) Transformation: The integration of a physics-derived or domain-specific transformation equation into a Transformer architecture represents a reusable design pattern for spatiotemporal forecasting.

Rejected Candidates

  • [concept] ASTEROID Framework (asteroid-framework) - paper_local: This is an application-specific framework name rather than a reusable architectural mechanism.

Metadata & Links

url
https://arxiv.org/abs/2606.17668
paper_id
2606.17668
paper_source
openalex
domain
time-series
tags
transformerspatiotemporal-modelingforecastingmolecular-dynamics
architectures
datasets
concept_slugs
spatiotemporal-information-sti-transformation
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-06-19T06:46:39Z