protein-similarity-search

name	protein-similarity-search
description	Find homologous protein sequences with MMseqs2 (fast, default) or BLAST (comprehensive fallback) from a sequence or FASTA. Use to find sequence homologs or infer function by sequence similarity — not by structural similarity. Public APIs, no credential needed.
license	Apache-2.0

Protein Similarity Search (Codex-native)

Codex-native adaptation of Google DeepMind's protein-sequence-similarity-search skill (science-skills, Apache-2.0). Uses the public API directly through Codex's own tools.

Gates (ask before proceeding)

• Terms notice: on first use in a workspace, tell the user to review the source's terms of use and record the notice in provenance.
• Network / install: read-only public access; be polite about rate limits. Write large responses to a file and parse them; if you write helper code, run it via uv run, never bare python3. Ask before installs or large downloads.

Endpoints

• ColabFold MMseqs2 https://api.colabfold.com
• EBI NCBI BLAST REST https://www.ebi.ac.uk/Tools/services/rest/ncbiblast

Workflow

1. Submit the query sequence to MMseqs2 first; fall back to EBI BLAST if needed.
2. Report top homologs with identity, coverage, and E-value, and the inferred function.

• Provenance & review: record queries, endpoints, and cited identifiers with $science-provenance; check claims with $science-review before presenting.

Boundaries

• Sequence similarity suggests, not proves, homology/function; for structural homologs use $cx-foldseek-structural-search.
• Cite only hits actually returned.