skills.rest
mims-harvard
mims-harvardOFFICIAL
Explore all skills by this author →
Website𝕏@marinkazitnikGitHub

tooluniverse-protein-sae-variant-interpretation

Interprets missense variants by comparing ESMC-6B sparse autoencoder feature activations between reference and mutant protein sequences.

1.7k|254|Updated Mar 3, 2025
Search Tags:#uniprot#variant interpretation#sparse autoencoder#protein language model#loss-of-function#esmc-6b#missense variant
One-click install
npx skills add https://github.com/mims-harvard/ToolUniverse --skill tooluniverse-protein-sae-variant-interpretation
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: tooluniverse-protein-sae-variant-interpretation
Source: https://github.com/mims-harvard/ToolUniverse/tree/main/plugins/tooluniverse/skills/tooluniverse-protein-sae-variant-interpretation
Command: npx skills add https://github.com/mims-harvard/ToolUniverse --skill tooluniverse-protein-sae-variant-interpretation

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires esm.

What problem does it solve?

Pathogenicity scores like AlphaMissense or ClinVar tell you a missense variant is damaging but not why. This Skill provides a mechanistic explanation by showing which interpretable SAE features (catalytic, ligand-binding, PTM, structural motif, domain) are lost or gained at the mutation site in the ESMC-6B protein language model.

Core Features & Use Cases

  • Mechanistic Variant Interpretation: Compares reference vs mutant SAE feature activations in a window around the mutation to identify which biological capabilities are disrupted.
  • Composite One-Call Analysis: Uses ESM_explain_variant_mechanism to get disruption deltas, per-feature biological category labels, and a mechanism summary in a single call, with batch support for saturation mutagenesis.
  • Cross-Validation Workflow: Integrates gnomAD population frequency, AlphaMissense scores, ClinVar evidence, AlphaFold structure, and UniProt annotations to build a multi-layer confidence assessment.
  • Use Case: A researcher finds TP53 R175H is flagged damaging by AlphaMissense. This Skill reveals that catalytic and ligand-binding SAE features are lost at position 175, supporting a catalytic loss-of-function mechanism for the report.

Quick Start

Ask the agent to explain why the variant R175H in TP53 is loss-of-function using SAE feature analysis.

Frequently Asked Questions about tooluniverse-protein-sae-variant-interpretation

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

FAQPage Schema
How do I interpret why a missense variant is loss-of-function?▼

Provide a UniProt accession or gene symbol plus a variant like R175H. The skill fetches the canonical sequence, builds the mutant, and compares SAE feature activations to show which functional features (catalytic, binding, PTM) are lost or gained.

What is the difference between SAE variant interpretation and AlphaMissense?▼

AlphaMissense predicts whether a variant is damaging but gives no mechanism. SAE feature analysis explains which biological capability breaks, such as a catalytic site or ligand-binding pocket, complementing rather than replacing pathogenicity scores.

What API key and package does the ESMC SAE analysis require?▼

You need an ESM_API_KEY from EvolutionaryScale Forge and the esm package installed from the upstream GitHub branch, since the PyPI release does not yet include SAEConfig. Forge outputs are licensed for non-commercial research use.

Can I analyze multiple variants at the same residue in one batch?▼

Yes, ESM_score_variant_sae_batch accepts a list of variants, such as all 19 alternate amino acids at one position for saturation mutagenesis. It reuses the reference encoding, so 19 variants cost 20 Forge calls instead of 38.

What are the limitations of SAE-based variant interpretation?▼

SAE features are learned patterns, not ground-truth annotations, and the analysis covers only a ±8 residue window around the mutation. It misses long-range allosteric effects, uses only the canonical UniProt isoform, and some features lack biological category labels.

Related Skills

Explore More Education & Research Skills→
DrugClawDrugClaw
126

bio-db-tools

Query public biology databases and APIs for protein, variant, and pathway data.

Official
Intermediate
nahisahonahisaho
3

scientific-variant-interpretation

Automate ACMG-aligned clinical interpretation of genetic variants and generate standardized reports.

Community
Advanced
stellaromicsstellaromics

bio-clinical-databases-variant-prioritization

Prioritize disease-causing variants from annotated variant lists using ClinVar, gnomAD, CADD, and REVEL scores.

Official
Advanced
mims-harvardmims-harvard
1.6k

Clinical Variant Interpreter

Integrate population, ClinVar, ClinGen, OMIM, and literature data to classify genetic variants with ACMG criteria.

Official
Advanced