mlxmolkit is a GPU-accelerated molecular toolkit using MLX on Apple Silicon, it is a port of nvMolKit that uses Cuda.
It now includes
| Area | What it does | Entry point |
|---|---|---|
| Conformers | Drop-in for RDKit EmbedMolecules: DG (4D) → ETK (3D) → MMFF94, all on Metal. 8 ETKDG variants. N×k parallel | generate_conformers_nk |
| Clustering | Morgan FP → Tanimoto → Butina, at 150k+ molecules with divide-and-conquer memory | butina_tanimoto_mlx |
| NDDO semi-empirical | MNDO, RM1, AM1, PM3, PM6, PM7 and corrected variants; scalar and batched SCF, gradients/geometry optimization | mlxmolkit.nddo |
| xTB | GFN0/1/2 and g-xTB energies, analytical gradients, ANCOPT geometry optimization, ALPB water solvation | mlxmolkit.xtb |
| COSMO / COSMO-RS | σ-profiles, σ-potentials, activity coefficients, solubility in solvent mixtures | mlxmolkit.xtb (σ), mlxmolkit.cosmo(ddCOSMO) |
| Similarity & descriptors | ERG fingerprints, dense cosine, CHEESE embeddings, Connolly surfaces, dipole atom features | top-level exports |
Yo install
conda install -c conda-forge rdkit
pip install mlxmolkit-rdkit