molmim-nim

To generate novel small-molecule structures from a seed SMILES string, you can use CMA-ES guided latent space sampling. This approach allows you to create a library of analogs around a known molecule for drug discovery workflows.

Property-guided optimization improves lead candidates by using the CMA-ES algorithm to maximize desired molecular properties like QED or plogP. This systematically navigates the chemical latent space to filter and enhance drug-likeness.

Yes, you need both RDKit and NVIDIA BioNeMo NIM API access for molecular embedding tasks. RDKit handles local chemical structure validation, while the MolMIM NIM container or hosted endpoints process the latent space manipulations.

Yes, you can manipulate hidden states and decode molecules locally using BioNeMo. The skill provides access to advanced local-only endpoints for embedding, hidden-state manipulation, and decoding of molecular structures.

The best way to optimize small molecules for drug-likeness before docking simulations is to apply CMA-ES guided property optimization. This generates novel analogs from a seed molecule, filters them for drug-likeness, and prepares them for downstream analysis.