Quick overview

subseq.bio is a minimal web + API service for running protein design/analysis and related workloads. It hosts pre-configured, open-source models and programs such as RFdiffusion3, BoltzGen, AlphaFold, and others, in a simple to use interface.

Backstory

I’ve been obsessed with molecular nanotechnology for a long time. Just before ChatGPT was announced I was following the RFdiffusion work from Baker Lab at the Institute for Protein Design and it felt like a clear inflection point for practical synthetic protein generation.

Since then there’s been a steady stream of new ML models in this space, so I put together a system for composing and running them through a consistent and programmatic interface: subseq.bio.

Technical bits

- All programs use open-source code and weights. (no licensing restrictions and good for reproducibility) - API-first: anything you can do in the UI is available via the API. - Focused on synthetic protein design and related workloads. - Jobs are charged per use (no subscriptions), but new sign-ins get free credits so you can try things out.

I also recently added an MCP server for AI agent integration: https://subseq.bio/mcp

There’s no OAuth yet, but it works with an API key env var; you can grab a key from the site. Example codex config:

  export SUBSEQ_API_KEY=<subseq_api_key>
  codex mcp add subseq --url <subseq_mcp_url> --bearer-token-env-var SUBSEQ_API_KEY

For a visual overview, here’s a short video demo of a BoltzGen binder run on an AlphaFold2 output in the web UI: https://x.com/0xCF88/status/1995994854585696515

I know this is pretty niche and aimed at people already doing protein design / structure prediction, but I’d love to answer questions and to read any feedback.