Compute ESM2 embeddings and pseudo-log-likelihood scores for protein sequences.
npx skills add https://github.com/junior1p/ProteinClaw --skill esm-junior1p
Please help me install this Agent Skill. Skill: esm Source: https://github.com/junior1p/ProteinClaw/tree/main/skills/esm Command: npx skills add https://github.com/junior1p/ProteinClaw --skill esm-junior1p
Provides high-quality protein sequence embeddings and pseudo-log-likelihood scoring to assess sequence plausibility, prioritize design candidates, and enable embedding-based analyses without manual model engineering.
Compute embeddings and PLL scores for sequences.fasta and save embeddings to embeddings/ using the esm skill.
High-intent search queries and answers about installing and using this skill.
Pseudo-log-likelihood scoring in protein design calculates PLL scores using ESM2 to rank sequence naturalness, allowing you to filter implausible candidates and prioritize likely variants for structure prediction.
You score protein variants for plausibility by computing pseudo-log-likelihood scores directly from pretrained ESM2 models, which ranks sequence naturalness and filters out unlikely designs without manual model engineering.
Yes, you need a PyTorch-capable environment with sufficient GPU memory and access to pretrained ESM2 models to produce per-residue and pooled embeddings for large-scale sequence screening.
Yes, you can use this skill for large-scale peptide sequence screening to remove implausible sequences, cluster candidates by embedding similarity, and prioritize variants for structure prediction.
The best way to choose an ESM2 model size for protein sequence scoring is to balance speed and quality, using the provided GPU and runtime guidance for campaign planning to optimize throughput.
zongtingweiScore protein sequences and generate embeddings using the ESM2 model.
jimmc414Generate protein sequences and extract embeddings using ESM3 and ESM C.
adaptyvbioCompute protein embeddings and PLL scores using the ESM2 language model.
omar-A-hassanGenerate per-residue and per-sequence embeddings with ESM-2 and predict structures via ESMFold.
