protein-lm

name	protein-lm
description	Protein language model embeddings, diversity sampling, and mutation prediction using ESM or GenSLM
metadata	{"openclaw":{"requires":{"env":["OPENAI_API_KEY"],"bins":["python3"],"pip":["torch","transformers","numpy","scipy","faiss-cpu"]},"primaryEnv":"OPENAI_API_KEY"}}

Protein Language Model — ESM / GenSLM

Generate protein embeddings, cluster sequences for diversity sampling, and predict mutation effects using large protein language models.

Capabilities

• Sequence embeddings: Dense vector representations via GenSLM-ESMc or ESM-2
• Diversity sampling: FAISS clustering + Boltzmann importance sampling over embedding space
• Mutation effect prediction: Log-likelihood ratio scoring of single-point mutations

Scripts

scripts/run_embedding.py

Compute embeddings for protein sequences from a FASTA file or inline list.

python skills/protein-lm/scripts/run_embedding.py \
  --sequences MKTL... MGSS... \
  --model esm2 \
  --output-dir ./artifacts

scripts/run_sampling.py

Cluster pre-computed embeddings and sample representative sequences.

python skills/protein-lm/scripts/run_sampling.py \
  --embeddings-json ./artifacts/embeddings.json \
  --n-clusters 100 \
  --total-samples 500 \
  --output-dir ./artifacts

Parameters

• sequences: Amino acid sequences (FASTA file or inline)
• model: Model backend — esm2 (default) or genslm
• task: Task type — embedding, sample, mutation_effect
• n_clusters: Number of FAISS k-means clusters (default: 100)
• total_samples: Target number of sampled sequences (default: 500)
• mutations: Mutations for scoring (e.g., A50G L100F)
• alpha: Threshold ratio for mutation suggestions (default: 1.0)