Problem & Motivation
Molecular dynamics (MD) simulations have become a cornerstone of computational biology, with thousands of trajectories now publicly available across multiple repositories. Yet despite this data deluge, researchers face a fundamental challenge: similarity in MD is poorly defined. Unlike static structural biology where RMSD or TM-score provide clear metrics, MD similarity must account for both structural conformation and temporal dynamics—two dimensions that current tools fail to integrate.
The existing landscape is fragmented: repositories like Zenodo, MDRepo, Figshare, and OSF organize data primarily by metadata (author, date, DOI), offering no native support for content-based retrieval. There is no unified similarity engine that can answer questions like “find simulations with similar dynamic behavior to my trajectory” or “locate structurally related proteins beyond sequence homology.” This gap forces scientists to manually sift through terabytes of data, often missing relevant simulations simply because they weren't tagged with the right keywords.
Our Ambition
MD Find aims to bridge this divide by introducing multi-modal similarity search for molecular dynamics. We combine standardized metadata extraction (via GROMACS MetaDump) with vector embeddings of both static protein structures (using Progres) and trajectory dynamics (via kinetic-feature preservation networks).
The result is a searchable index spanning ~5,000 simulations from six public repositories, accessible through a web interface that supports browsing, text search, structure-based query, and dynamics-based query. Our ambition extends beyond tool-building: we seek to establish a community standard for how MD data should be indexed, queried, and shared, enabling the discovery and cross-system integration that was previously impossible.
Contact
If you have any questions, feedback, or inquiries, reach out to us at alphafind@ics.muni.cz.
Funding
This project was supported by the CESNET Development Fund (grant No. 776/2025).
Source code
The source code is available on GitLab: https://gitlab.ics.muni.cz/md-find-public/md-find.
Data availability
The underlying data used in this project are available upon request. Contact us if you're interested.