About MolCrystalFlow
MolCrystalFlow is a web interface for molecular crystal candidate generation. Given a molecular conformer as input, the system generates candidate crystal packings, filters physically invalid outputs, ranks accepted candidates by learned interatomic-potential energy, and returns downloadable structure files with interactive 3D previews.
Developed at the University of Florida.
Input
Molecular conformer
Standard XYZ format with atomic coordinates
Output
Ranked crystal candidates
Candidate packings for supported Z values
Method
MolCrystalFlow targets the crystal packing generation stage of crystal structure prediction (CSP). Unlike composition-first approaches, MolCrystalFlow is conformer-first: the user provides a molecular conformer in XYZ format, and the model generates candidate crystal packings at specified Z values (molecules per unit cell).
The current workflow applies geometry validation and periodic molecular clash filtering before ranking accepted candidates with UMA-s-1p2 energies after 10 rigid-body optimization steps. Energy is used for ranking, not as an automatic cutoff.
Outputs are computational candidates intended for inspection and downstream scientific validation, not experimentally confirmed crystal structures.
Model
Generator
MolCrystalFlow CCDC v2
ccdc-v2-epoch436
Ranking
UMA-s-1p2
Rigid-body optimization followed by potential-energy ranking
Supported outputs
Z = 2 and Z = 4 candidate packings
Downloadable XYZ files with lattice metadata and interactive previews