NIM for MSA Search
Type: Microservice Tags: NVIDIA, NIM, BioNeMo, MSA, multiple sequence alignment, MMSeqs2, protein structure prediction, AlphaFold, OpenFold, Boltz Related: NVIDIA-BioNeMo, NVIDIA-NIM, NIM-for-AlphaFold2, NIM-for-AlphaFold2-Multimer, NIM-for-OpenFold2, NIM-for-OpenFold3, NIM-for-Boltz2, NIM-for-ProteinMPNN, NVIDIA-Clara, NVIDIA-AI-Enterprise, NVIDIA-CUDA, NGC Sources: https://docs.nvidia.com/nim/bionemo/msa-search/latest/overview.html Last Updated: 2026-04-29
Summary
NIM for MSA Search is NVIDIA’s BioNeMo NIM for GPU-accelerated multiple sequence alignment of query amino-acid sequences against protein sequence databases. Current NVIDIA docs describe it as using GPU-accelerated MMSeqs2 and providing inputs for structure prediction models such as AlphaFold2, OpenFold, Boltz, and multimer workflows.
Detail
Purpose
Multiple sequence alignment helps structure prediction models use evolutionary and homologous-sequence information. MSA Search NIM makes that search/align step deployable as a GPU-accelerated microservice rather than a separate CPU-heavy preprocessing pipeline.
Current scope
- Searches protein sequence databases for similar sequences and aligns related sequences.
- Supports AlphaFold2-style monomer search with a single-pass search per database.
- Supports a ColabFold-style cascaded search process for higher sensitivity and throughput.
- Supports paired MSA search for protein complexes by pairing homologous chain sequences by species.
- Supports structural template search against structural databases such as PDB70 and returns template hits plus MSA alignments.
- Uses GPU-accelerated MMSeqs2 for improved latency and throughput.
NVIDIA context
MSA Search is infrastructure for the BioNeMo structure prediction stack. It connects directly to NIM-for-AlphaFold2, NIM-for-AlphaFold2-Multimer, NIM-for-OpenFold2, NIM-for-OpenFold3, and NIM-for-Boltz2 rather than being a standalone molecular design model.
Connections
- NIM-for-AlphaFold2 and NIM-for-AlphaFold2-Multimer - MSA inputs improve structure prediction and multimer modeling.
- NIM-for-OpenFold2 and NIM-for-OpenFold3 - OpenFold-style NIMs can use MSA/template context.
- NIM-for-Boltz2 - complex prediction can use homologous sequence and template context.
- NIM-for-ProteinMPNN - protein design workflows can combine structure generation/design with sequence context.
- NVIDIA-BioNeMo and NVIDIA-Clara - life-sciences platform context.
- NVIDIA-AI-Enterprise, NVIDIA-CUDA, and NGC - deployment, acceleration, and distribution context.
Source Excerpts
- NVIDIA docs describe MSA Search NIM as supporting GPU-accelerated MSA against protein sequence databases.
- The current docs list AlphaFold2 search, ColabFold search, paired MSA search, structural template search, and GPU-accelerated MMSeqs2.