Protein Language Model Function Analysis Starter

Use this skill to validate protein FASTA inputs, extract deterministic smoke-safe embeddings, and run a reusable sequence-to-function triage flow that can later be swapped onto real ESM-2 or ProtT5 backends.

What This Skill Does

• reads protein sequences from FASTA and rejects unsupported residue symbols
• emits per-sequence embeddings to a TSV contract
• runs centroid-based function analysis when labels are supplied
• reports nearest-neighbor structure in embedding space
• preserves a stable CLI for real transformers backends such as ESM-2 and ProtT5

When To Use It

• when you need a concrete local skill for protein-embeddings and sequence-to-function-modeling
• when you want a deterministic smoke path before moving to GPU-backed protein language model inference
• when you need a compact handoff artifact for downstream DeepFRI, annotation, or benchmarking work

Run

python3 skills/proteomics/protein-language-model-function-analysis-starter/scripts/run_protein_language_model_function_analysis.py \
  --input skills/proteomics/protein-language-model-function-analysis-starter/examples/toy_sequences.fasta \
  --labels skills/proteomics/protein-language-model-function-analysis-starter/examples/toy_labels.tsv \
  --config skills/proteomics/protein-language-model-function-analysis-starter/examples/analysis_config.json \
  --embeddings-out scratch/protein-lm/toy_embeddings.tsv \
  --summary-out scratch/protein-lm/toy_summary.json

Notes

• The default mock backend is intentionally deterministic and test-friendly. It preserves the same file contract as a real protein language model run.
• For live model inference, switch the backend to transformers and point model_id at an ESM-2 or ProtT5 checkpoint. See refs.md for canonical sources and formatting notes.
• The bundled toy labels are illustrative and suitable only for smoke testing or pipeline scaffolding, not biological claims.