NIM for GenMol
Type: Microservice Tags: NVIDIA, NIM, BioNeMo, GenMol, fragment-based small molecule generation, SAFE, masked diffusion, drug discovery Related: NVIDIA-BioNeMo, NVIDIA-NIM, NIM-for-MolMIM, NIM-for-DiffDock, NIM-for-ALCHEMI-Batched-Geometry-Relaxation, NVIDIA-Clara, NVIDIA-AI-Enterprise, TensorRT, Triton-Inference-Server, NGC Sources: https://docs.nvidia.com/nim/bionemo/genmol/latest/overview.html Last Updated: 2026-04-29
Summary
NIM for GenMol is NVIDIA’s BioNeMo NIM for fragment-based small molecule generation. Current NVIDIA docs describe GenMol as a masked diffusion model trained on SAFE representations, allowing users to specify fixed fragments, attachment positions, fragment lengths, and partial or full generation schemas.
Detail
Purpose
GenMol supports molecular design workflows where chemists want to generate molecules from fragments rather than only sampling from a broad latent space. It is useful for scaffold decoration, motif extension, linker design, hit generation, and lead optimization.
Current scope
- Uses SAFE-formatted molecular sequences for flexible fragment-level design.
- Performs masked-token generation with a Transformer/BERT-style network and iterative unmasking process.
- Supports de novo generation of valid molecular sequences at requested lengths.
- Supports conditioned generation, including motif extension, scaffold decoration, superstructure generation, and linker design.
- Supports molecule optimization workflows with oracle methods for hit generation and lead optimization.
- Exposes NIM-style HTTP/OpenAPI request patterns for self-hosted or hosted inference workflows.
NVIDIA context
GenMol is the current Fragment-Based Small Molecule Generation NIM in the NVIDIA NIM index. It complements NIM-for-MolMIM for controlled small molecule generation and NIM-for-DiffDock for downstream protein-ligand pose prediction.
Connections
- NVIDIA-BioNeMo - molecular generation and drug-discovery platform context.
- NIM-for-MolMIM - controlled small molecule generation NIM.
- NIM-for-DiffDock - docking/pose-prediction NIM for generated molecules.
- NIM-for-ALCHEMI-Batched-Geometry-Relaxation - atomistic relaxation NIM for molecular or materials candidates.
- NVIDIA-Clara - broader NVIDIA healthcare and life-sciences graph.
- NVIDIA-AI-Enterprise - enterprise deployment and support context.
- TensorRT, Triton-Inference-Server, and NGC - optimized inference, serving, and container distribution context.
Source Excerpts
- NVIDIA docs describe GenMol as a masked diffusion model trained on SAFE representations for fragment-based molecule generation.
- The current docs list de novo generation, conditioned generation, linker design, and oracle-guided optimization use cases.