tooluniverse-protein-structure-prediction

Predicts protein 3D structures from sequence using ESMFold, AlphaFold, and RCSB experimental data.

1.7k|254|Updated Mar 3, 2025
One-click install
npx skills add https://github.com/mims-harvard/ToolUniverse --skill tooluniverse-protein-structure-prediction
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: tooluniverse-protein-structure-prediction
Source: https://github.com/mims-harvard/ToolUniverse/tree/main/plugins/tooluniverse/skills/tooluniverse-protein-structure-prediction
Command: npx skills add https://github.com/mims-harvard/ToolUniverse --skill tooluniverse-protein-structure-prediction

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Researchers often need a 3D protein structure when no experimental structure exists, and manually coordinating prediction tools, confidence scoring, and variant analysis is slow and error-prone. This Skill runs an end-to-end workflow that predicts structures from sequence, benchmarks them against experimental data, and interprets variant impact.

Core Features & Use Cases

  • De Novo Structure Prediction: Runs ESMFold on any amino acid sequence (up to ~800 residues) and reports per-residue pLDDT and pTM confidence scores.
  • AlphaFold & Experimental Comparison: Retrieves precomputed AlphaFold models by UniProt accession and searches RCSB PDB for experimental structures to validate predictions.
  • Variant Impact Assessment: Uses ProtVar to map mutations like "P04637 R175H" to structural and functional context, with tiered evidence grading.
  • Use Case: Given a novel protein sequence, the Skill computes physicochemical properties with ProtParam, predicts the fold with ESMFold, cross-checks against AlphaFold and PDB, and delivers a structured report with confidence maps and recommendations.

Quick Start

Predict the structure of this protein sequence and tell me which regions are low confidence: MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSH.

Frequently Asked Questions about tooluniverse-protein-structure-prediction

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I predict a protein structure from an amino acid sequence?▼

Provide the single-letter amino acid sequence to ESMFold_predict_structure, which returns a PDB-format structure with per-residue pLDDT confidence scores and a pTM global fold score. Sequences up to about 800 residues are supported.

ESMFold vs AlphaFold: which should I use for structure prediction?▼

ESMFold works directly on any sequence without a database lookup and is faster for novel proteins. AlphaFold uses multiple sequence alignments and typically gives higher accuracy for well-conserved proteins with a UniProt accession, so use it as the reference when available.

What does a pLDDT score below 50 mean in AlphaFold predictions?▼

A pLDDT below 50 indicates very low confidence, usually corresponding to intrinsically disordered regions rather than a folded structure. These regions should not be interpreted as having a defined 3D conformation.

Can ESMFold predict structures for sequences longer than 800 residues?▼

Sequences over roughly 800 residues may fail or produce lower-quality predictions with ESMFold. The recommended fallback is to retrieve the precomputed AlphaFold model using the protein's UniProt accession instead.

How do I assess whether a mutation affects protein structure?▼

Use ProtVar_map_variant with notation like "P04637 R175H" to resolve the position, then ProtVar_get_function to get domain context, conservation, and pathogenicity predictions. Residues in active sites or buried hydrophobic cores typically have higher structural impact.

Can this workflow predict protein complexes or multimers?▼

No, both ESMFold and standard AlphaFold predict single-chain monomer structures only. Complex prediction requires AlphaFold-Multimer, which is not available through these tools.