NVIDIA BioNeMo / NIM
GPU-accelerated inference and fine-tuning for protein structure, protein design, docking, and genomics foundation models via NIM microservices.
Overview
NVIDIA BioNeMo is where teams run GPU-accelerated inference on the biological foundation models that matter for drug discovery, packaging each as a NIM microservice: a containerized, hardware-optimized API endpoint. If you are searching for a protein structure prediction API, a hosted protein design model, or a genomics model you can call over REST, BioNeMo bundles them behind one production-grade interface. The companion BioNeMo Framework adds training and fine-tuning for supported model families, so the same platform covers both zero-shot inference and domain adaptation.
What you can run on NVIDIA BioNeMo / NIM
Protein structure prediction spans AlphaFold 2 and AlphaFold-Multimer for monomers and complexes, OpenFold as an open reimplementation, ESMFold for single-sequence folding, and Boltz-2 for all-atom complex and affinity prediction. For protein design, RFdiffusion generates novel backbones and ProteinMPNN handles inverse folding and sequence design. ESM-2 delivers protein language-model embeddings and variant scoring, while Evo 2 extends the catalog into genomics with long-context DNA modeling. Together these cover structure prediction, docking and affinity, de novo design, and sequence-to-function tasks across proteins and nucleic acids.
Running inference on NVIDIA BioNeMo / NIM
Every model is served as a NIM: call it as a managed hosted endpoint on NVIDIA's cloud, or pull the same container and self-host it on your own GPU infrastructure for data-controlled, on-prem deployment. Access is through REST APIs and OpenAI-style client calls, making the endpoints easy to wire into pipelines and agent workflows. OpenFold and ESM-2/ESMFold additionally support fine-tuning through the BioNeMo Framework, so you can adapt those families to proprietary sequences and structures. This split suits enterprise and techbio teams that need scalable, accelerated inference with the option to keep sensitive data inside their own environment, rather than one-off notebook runs.
Run inference on NVIDIA BioNeMo / NIM (8)
Protein structure prediction model that folds amino acid sequences into 3D structures with atomic accuracy, scoring a median GDT of 92.4 at CASP14.
Meta AI's family of protein language models scaled to 15B parameters, paired with ESMFold for fast, alignment-free atomic-level structure prediction.
Protein complex structure prediction model extending AlphaFold 2 with paired MSA processing and ipTM scoring for multi-chain, multimeric assemblies.
Message passing neural network for fixed-backbone protein sequence design. Achieves 52.4% native sequence recovery, far surpassing Rosetta's 32.9%.
De novo protein design diffusion model that generates backbone structures conditioned on binding targets, symmetry constraints, and functional motifs.
Open model that jointly predicts biomolecular structure and small-molecule binding affinity, approaching FEP+ accuracy in seconds on a single GPU.
Trainable, open-source reimplementation of AlphaFold2 for protein structure prediction that matches its accuracy and runs 3-5x faster.
Genomic foundation model trained on 9.3 trillion DNA base pairs across all domains of life, with 40B parameters and a 1-million-token context.
Fine-tune on NVIDIA BioNeMo / NIM (2)
Meta AI's family of protein language models scaled to 15B parameters, paired with ESMFold for fast, alignment-free atomic-level structure prediction.
Trainable, open-source reimplementation of AlphaFold2 for protein structure prediction that matches its accuracy and runs 3-5x faster.