🧬

esm

Name: esm
Author: K-Dense-AI

Safe ⚡ Contains scripts🌐 Network access⚙️ External commands📁 Filesystem access

Generate and design proteins with ESM

Also available from: davila7

Protein engineering requires analyzing and designing novel protein sequences. ESM provides state-of-the-art protein language models for generating sequences, predicting structures, and creating embeddings for downstream analysis. Use ESM3 for generative tasks across sequence, structure, and function, or ESM C for efficient representation learning.

Supports: Claude Codex Code(CC)

📊 71 Adequate

Download the skill ZIP

Upload in Claude

Go to Settings → Capabilities → Skills → Upload skill

Toggle on and start using

Test it

Using "esm". Generate a complete GFP variant using ESM3. The sequence should have the chromophore motif 'SYG' at positions 65-67.

Expected outcome:

Generated GFP variant sequence (238 amino acids) with chromophore motif preserved. The model produced a sequence with characteristic GFP barrel structure. Sequence can be validated through structure prediction.

Using "esm". Extract embeddings for these three protein sequences and compute pairwise similarity.

Expected outcome:

Embedding vectors (1280 dimensions for ESM C-300M). Pairwise cosine similarities: 0.72 between sequences A and B, 0.45 between A and C, 0.68 between B and C.

Security Audit

Safe

v5 • 1/21/2026

All 368 static findings are false positives. The scanner incorrectly flagged markdown documentation patterns. The skill provides documentation for legitimate protein language models from EvolutionaryScale. All code examples are standard scientific workflows for protein engineering. Python f-strings with underscores (protein masks), MD5 for cache keys, and ML terminology were misclassified as security issues.

Files scanned

6,067

Lines analyzed

findings

Total audits

Risk Factors

⚡ Contains scripts (5)

evaluation.json:51 evaluation.json:54 references/esm-c-api.md:338 skill-report.json:1208 skill-report.json:1209

🌐 Network access (21)

⚙️ External commands (188)

📁 Filesystem access (13)

references/esm-c-api.md:462 references/esm3-api.md:55 references/esm3-api.md:191 references/forge-api.md:328 references/forge-api.md:463 references/forge-api.md:438 references/workflows.md:515 references/workflows.md:92 references/workflows.md:95 references/workflows.md:278 references/workflows.md:283 references/workflows.md:436 references/workflows.md:634

Audited by: claude View Audit History →

Quality Score

Architecture

100

Maintainability

Content

Community

100

Security

Spec Compliance

What You Can Build

Design novel fluorescent proteins

Generate GFP variants with desired properties using ESM3 function conditioning to target specific functional domains while exploring sequence space.

Optimize enzyme sequences

Use inverse folding to redesign enzyme sequences that maintain structural integrity while improving stability or activity for industrial applications.

Screen protein variants with embeddings

Generate embeddings for protein libraries and use clustering to identify promising candidates before experimental testing.

Try These Prompts

Generate protein with masked positions

Generate a complete protein sequence by filling in the masked positions. The input sequence uses '_' to represent masked residues that need to be generated. Use ESM3 with the sequence track to complete the protein: {partial_sequence}

Predict structure from sequence

Use ESM3 to predict the 3D structure of this protein sequence. Generate structure coordinates using the structure track. Return the PDB-formatted coordinates: {protein_sequence}

Design sequence for target structure

Perform inverse folding to design a protein sequence that folds into the target structure provided as PDB. Use ESM3 to generate a sequence that is compatible with the 3D coordinates: {pdb_content}

Extract protein embeddings

Generate high-quality embeddings for this protein sequence using ESM C. Return the embedding vector for downstream analysis or similarity comparison: {protein_sequence}

Best Practices

Start with smaller open models for prototyping before scaling to larger Forge API models for production quality
Use chain-of-thought generation for complex protein designs by iterating through sequence, structure, and function tracks
Validate generated sequences with structure prediction and follow Responsible Biodesign Framework guidelines

Avoid

Do not skip experimental validation of computationally designed proteins
Do not use generated proteins directly in clinical applications without proper testing
Do not ignore biosafety implications when designing proteins with novel functions

Frequently Asked Questions

What is the difference between ESM3 and ESM C models?

ESM3 is a generative model for creating novel proteins across sequence, structure, and function. ESM C is an encoder-only model focused on generating high-quality embeddings for analysis, classification, and similarity comparison tasks.

Can I use ESM3 locally or do I need the Forge API?

The smallest ESM3 model (1.4B parameters) is available as open weights for local execution. Larger models (7B and 98B) require the Forge API for inference. ESM C models are available locally at various sizes.

How long can protein sequences be for ESM processing?

Local models are limited by GPU memory, typically handling sequences up to 2000 residues. The Forge API can process longer sequences with its optimized infrastructure.

What is inverse folding and when should I use it?

Inverse folding predicts a protein sequence that folds into a given 3D structure. Use it when you have a designed structure and need a sequence that will adopt that fold, or when adapting proteins to new scaffolds.

How do I validate proteins generated by ESM?

Validate generated proteins using structure prediction to check fold validity, checking for known protein family motifs, and following experimental protocols. Always consult the Responsible Biodesign Framework.

What are generation tracks in ESM3?

ESM3 supports three generation tracks: sequence (amino acids), structure (3D coordinates), and function (GO annotations). These can be used independently or together for multimodal protein design.

Developer Details

Author

K-Dense-AI

License

MIT license

Repository

https://github.com/K-Dense-AI/claude-scientific-skills/tree/main/scientific-skills/esm

Ref

main

File structure

📁 references/

📄 esm-c-api.md

📄 esm3-api.md

📄 forge-api.md

📄 workflows.md

📄 evaluation.json

📄 SKILL.md