msa-nim 0.3.0

CLI tool to generate A3M MSA alignments via NVIDIA NIM MSA-Search

msa-nim

A3M alignments for AlphaFold — no GPUs, no database downloads

---

Install

pip install msa-nim

Quick start

# 1. Install
pip install msa-nim

# 2. Put your .fasta files in a folder
cd my_proteins/

# 3. Run — it'll ask for your API key on first use and save it
msa-nim run

# Done. A3M files are in msa_results/

First run prompts for your NVIDIA NIM API key (free) and saves it to .msa-nim.env for future runs. Same MSA engine as ColabFold — just without the 1.4 TB database download.

From a PDB ID

No FASTA file needed:

msa-nim pdb 7DKF              # all chains
msa-nim pdb 7DKF --chain A    # specific chain
msa-nim pdb 7DKF 6HBB 1ABC    # multiple IDs

Options

msa-nim run /path/to/fastas       # custom input directory
msa-nim run -o my_output          # custom output directory
msa-nim run --resume               # retry crashed/interrupted run
msa-nim run -j 2                   # parallel jobs (paid API tier only)
msa-nim run --db PDB70_220313      # add structural templates

Note: The free NVIDIA tier rate-limits to ~1 request at a time. Use -j 1 (default) on free tier; -j 2+ only helps on paid plans.

FASTA format

One sequence per > record. Files must end in .fasta, .fa, or .faa:

>my_protein
MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQAPILSRVG...

>another_protein
MVHLTPEEKSAVTALWGKVNVDEVGGEALGRLLVVYPWTQR...

Output

msa_results/ contains {name}_{db}.a3m files — ready for AlphaFold, ColabFold, or OpenFold.

Troubleshooting

Problem	Fix
No .fasta files found	Use .fasta, .fa, or .faa extensions
422: Database not available	Check spelling: Uniref30_2302, colabfold_envdb_202108, PDB70_220313
Interrupted run	Re-run with --resume
Rate limiting (429)	Free tier is ~1 req/sec; use default -j 1

License

MIT — see LICENSE.

Acknowledgements

• MMseqs2-GPU — GPU-accelerated homology search
• ColabFold — MSA + structure prediction pipeline
• NVIDIA NIM — Cloud inference microservices