BioNeMo NIM Overview | NVIDIA BioNeMo NIM Documentation

NVIDIA BioNeMo NIMs are a collection of GPU-accelerated inference microservices that bring state-of-the-art AI models for biology and drug discovery behind simple, consistent APIs. Each NIM packages a leading model as a self-hosted, easily deployable container, and because they share a common interface, they can be chained together to build complete in silico drug discovery pipelines spanning structure prediction, molecular generation, protein design, genomic analysis, and molecular docking.

Use Cases

BioNeMo NIMs span structure prediction, molecular generation and design, genomics, and docking.

Structure Prediction

Use these NIMs when you need 3D structures or alignments from sequence:

AlphaFold2

Predict protein structures from amino acid sequences.

AlphaFold2-Multimer

Predict protein-complex structures from amino acid sequences.

OpenFold2

An open, PyTorch reimplementation of AlphaFold2 at parity accuracy with improved speed.

OpenFold3

Predicts complete biomolecular complexes of proteins, DNA, RNA, and small-molecule ligands.

Boltz-2

Predicts biomolecular complexes and their binding affinity.

MSA Search

Fast, GPU-accelerated multiple sequence alignment that the folding models rely on to improve accuracy.

Benefits: Shared HTTP APIs, GPU-accelerated inference, pipeline-ready MSA and folding outputs

Molecular Generation and Design

Use these NIMs when you need novel small molecules or protein backbones and sequences:

MolMIM

Generates novel small molecules through a clustered latent space for controlled, property-guided generation.

GenMol

Generates small molecules through fragment-based (SAFE) masked-diffusion generation.

RFdiffusion

Generates novel protein backbones.

RFD4-Proteina

Generates novel protein backbones and sequences for tasks such as motif scaffolding, binder design and antibody CDR design.

ProteinMPNN

Predicts amino acid sequences that fold into a given protein structure.

Benefits: Controlled generation, fragment and property constraints, protein design chaining

Genomics and Docking

Use these NIMs when you need genomic interpretation or protein–ligand poses:

Evo 2

A genomic foundation model that interprets and generates DNA across scales, from individual molecules to entire genomes.

DiffDock

Predicts protein-ligand binding poses for molecular docking.

Benefits: Sequence-to-function exploration and docking poses with confidence estimates