Multi-algorithm protein structure prediction using Modal serverless infrastructure.
Runs Boltz-2, Chai-1, Protenix, AlphaFold2.
Set up uv and modal (mac and linux)
curl -LsSf https://astral.sh/uv/install.sh | sh
uvx modal setupRun dev server or deploy to Modal
uvx modal serve foldism.py
uvx modal deploy foldism.pyRun basic tests on command line
echo -e ">Insulin\nGIVEQCCTSICSLYQLENYCN\n>InsulinB\nFVNQHLCGSHLVEALYLVCGERGFFYTPKT" > input.faa
uvx modal run foldism.py --input-faa input.faa
uvx modal run foldism.py --input-faa input.faa --algorithms chai1,boltz2
uvx modal run foldism.py --input-faa input.faa --no-use-msa # faster but lower quality- Protenix's msa server is sometimes very slow, and this can cause Protenix to run for hours, potentially
- Modal currently gives you $30 per month free to experiment with
- A run with all 4 methods can cost $1 or more depending on sequence length (GPUs cost $1-4 per hour)
- Results are cached, so re-running the same sequence is almost free
- If you
modal deploy, the URL is publicly available (anyone who finds it can use it) - If you want to require users to authenticate, set up a custom domain. I use cloudflare but there are many ways to do this.
MIT
