Tooluniverse Variant Interpretation

Clinical Variant Interpreter

Systematic variant interpretation using ToolUniverse - from raw variant calls to ACMG-classified clinical recommendations with structural impact analysis.

Triggers

Use this skill when users:

• Ask about variant interpretation, classification, or pathogenicity
• Have VCF data needing clinical annotation
• Need ACMG classification for variants
• Want structural impact analysis for missense variants

Key Principles

1. ACMG-Guided - Follow ACMG/AMP 2015 guidelines with explicit evidence codes 2. Structural Evidence - Use AlphaFold2 for novel structural impact analysis 3. Population Context - gnomAD frequencies with ancestry-specific data 4. Actionable Output - Clear recommendations, not just classifications 5. English-first queries - Always use English terms in tool calls; respond in user's language

---

LOOK UP, DON'T GUESS

When asked about a variant's significance, query ClinVar/gnomAD/CIViC FIRST. Never classify a variant without checking databases. When you're not sure about a fact, your first instinct should be to SEARCH for it using tools, not to reason harder from memory.

---

Workflow Overview

Phase 1: VARIANT IDENTITY        → Normalize HGVS, map gene/transcript/consequence
Phase 2: CLINICAL DATABASES       → ClinVar, gnomAD, OMIM, ClinGen, COSMIC, SpliceAI
Phase 2.5: REGULATORY CONTEXT     → ChIPAtlas, ENCODE (non-coding variants only)
Phase 3: COMPUTATIONAL PREDICTIONS → CADD, AlphaMissense, EVE, SIFT/PolyPhen
Phase 4: STRUCTURAL ANALYSIS      → PDB/AlphaFold2, domains, functional sites (VUS/novel)
Phase 4.5: EXPRESSION CONTEXT     → CELLxGENE, GTEx tissue expression
Phase 5: LITERATURE EVIDENCE      → PubMed, EuropePMC, BioRxiv, MedRxiv
Phase 6: ACMG CLASSIFICATION      → Evidence codes, classification, recommendations

---

Phase 1: Variant Identity

Tools: MyVariant_query_variants, EnsemblVar_get_variant_consequences, NCBIGene_search, VariantValidator_gene2transcripts, VariantValidator_validate_variant, Tark_get_mane_transcripts, Tark_get_transcript

VariantValidator_gene2transcripts: Look up MANE Select and MANE Plus Clinical transcripts for a gene. Use this to identify the correct canonical transcript before variant annotation.

• Parameters: gene_symbol (e.g. "TP53"), transcript_set ("mane" | "refseq" | "ensembl" | "all"), genome_build ("GRCh38" default)
• Returns: Array of {current_symbol, transcripts: [{reference, annotations: {mane_select, mane_plus_clinical}}]}
• Aliases: gene and gene_name also accepted for gene_symbol

Tark_get_mane_transcripts: Lightweight MANE Select / MANE Plus Clinical lookup from Ensembl Tark with the ENST↔RefSeq pairing (e.g. gene="BRCA2" → ENST00000380152.8 / NM_000059.4). Use as a quick cross-check of the canonical transcript namespace alongside VariantValidator, or to translate an ENST↔NM accession. Tark_get_transcript (param stable_id, e.g. "ENST00000380152") returns the archived transcript record (assembly, biotype, coordinates, per-release versions) when you need to resolve a specific transcript version.

VariantValidator_validate_variant: Validate HGVS variant descriptions and get normalized notation with genomic/transcript/protein consequences.

• Parameters: genome_build ("GRCh37" | "GRCh38"), variant_description (HGVS, e.g. "NM_007294.4:c.5266dup"), select_transcripts (transcript or "all")
• Returns: Validated HGVS, protein consequence, genomic coordinates, gene IDs

Capture: HGVS notation (c. and p.), gene symbol, canonical transcript (MANE Select via VariantValidator), consequence type, amino acid change, exon/intron location.

Phase 2: Clinical Databases

Tools: ClinVar_search_variants, gnomad_search_variants, gnomad_get_variant, OMIM_search, OMIM_get_entry, ClinGen_search_gene_validity, ClinGen_search_dosage_sensitivity, ClinGen_search_actionability, COSMIC_search_mutations, COSMIC_get_mutations_by_gene, DisGeNET_search_gene, DisGeNET_get_vda, SpliceAI_predict_splice, SpliceAI_get_max_delta, civic_get_variants_by_gene, civic_search_evidence_items, civic_search_assertions

gnomAD two-step workflow: gnomad_search_variants only accepts rsIDs or variant IDs (not gene names). Search by rsID first, then use the returned variant_id with gnomad_get_variant to get population allele frequencies.

>

CIViC: Use civic_search_genes(query="<gene_symbol>") to find the CIViC gene ID dynamically (do NOT rely on a hardcoded lookup table). Then use civic_get_variants_by_gene(gene_id=<id>) and civic_search_evidence_items for actionability details. If civic_search_genes returns no results, the gene may not be curated in CIViC — note this gap.

>

OncoKB note: Demo mode only supports BRAF, TP53, ROS1. For other genes, set ONCOKB_API_TOKEN environment variable.

Use SpliceAI for: intronic variants near splice sites, synonymous variants, exonic variants near splice junctions.

See CODE_PATTERNS.md for implementation details.

Phase 2.5: Regulatory Context (Non-Coding Only)

Apply for intronic (non-splice), promoter, UTR, or intergenic variants near disease genes.

Tools: ChIPAtlas_enrichment_analysis, ChIPAtlas_get_peak_data, ENCODE_search_experiments, ENCODE_get_experiment

Phase 2.9: Short-Circuit Check

Before full ACMG classification, check if the variant already has an expert panel classification in ClinVar. Use MyVariant_query_variants with the rsID or HGVS notation — the clinvar field in the response includes clinical significance, review status, and RCV records. If an expert panel has already classified the variant as Pathogenic or Benign, note this prominently and focus on confirming/contextualizing rather than de novo classification.

Phase 3: Computational Predictions

Primary approach: MyVariant_query_variants with fields=dbnsfp,clinvar,cadd,gnomad_genome retrieves 15+ predictor scores (SIFT, PolyPhen, CADD, REVEL, AlphaMissense, MetaRNN, FATHMM, GERP, PhyloP, etc.) in a single call. This is usually sufficient.

REVEL/AlphaMissense fallback: If MyVariant_query_variants returns no dbnsfp block, use the dedicated tool: 1. `MyVariant_get_pathogenicity_scores` (PREFERRED FALLBACK) — returns REVEL, AlphaMissense, SIFT, PolyPhen2, MetaRNN, GERP, PhyloP, and more in a single call with pre-configured dbnsfp fields. Input: variant_id (rsID or HGVS genomic). 2. CADD_get_variant_score (PHRED 0-99) — works for most variants 3. AlphaMissense_get_variant_score (0-1, needs UniProt ID) — missense only 4. EVE_get_variant_score (0-1) — missense only 5. EnsemblVEP_annotate_hgvs (VEP with colocated variants) — includes SIFT/PolyPhen 6. If REVEL is still unavailable, note this as a limitation and rely on CADD + SIFT + PolyPhen consensus. REVEL absence does not prevent classification.

Consensus: Run CADD (all variants) + AlphaMissense + EVE (missense). 2+ concordant damaging = strong PP3; 2+ concordant benign = strong BP4.

See ACMG_CLASSIFICATION.md for thresholds.

Phase 4: Structural Analysis (VUS/Novel Missense)

Tools: PDBe_get_uniprot_mappings, NvidiaNIM_alphafold2 (requires NVIDIA_API_KEY env var; free key at build.nvidia.com), alphafold_get_prediction (param: qualifier, e.g., UniProt accession), InterPro_get_protein_domains, UniProt_get_function_by_accession

Workflow: Get structure -> map residue -> assess domain/functional site -> predict destabilization.

AlphaFold size limitation: Very large proteins (>2,700 aa, e.g., BRCA2 at 3,418 aa) may not have AlphaFold predictions via the standard API. Fall back to published structural studies or PDBe_get_uniprot_mappings for experimental structures.

Phase 4.2: Mechanism of Effect (VUS missense, ESMC-6B SAE)

AlphaMissense / REVEL / CADD give a pathogenicity score but no mechanism. When you need to answer "how does this variant disrupt protein function" — e.g. for VUS write-ups, clinical reports, or to triangulate a discordant predictor consensus — use the ESMC-6B Sparse Autoencoder to identify which interpretable protein-language-model features the mutation disrupts.

One-call mechanism summary (recommended starting point):

mech = tu.tools.ESM_explain_variant_mechanism(
    sequence=wt_aa_sequence,   # full reference protein sequence
    position=600,              # 1-indexed
    ref_aa="V",
    alt_aa="E",
    top_k_features=5,          # describe top 5 lost + top 5 gained
)
# mech["data"]["mechanism_summary"] e.g.:
#   "Disrupted feature categories (lost): catalytic=2, ligand-binding=1;
#    Induced feature categories (gained): structural-stability=1"

Returns mechanism_summary, per-feature lost/gained tables, and category aggregates. Use the category aggregate to support or qualify the pathogenicity verdict in the report:

• catalytic / ligand-binding / ptm lost → mechanistic support for PP3
• secondary-structure / structural-stability gained on a stable WT region → mechanistic basis for "destabilizing" claim
• No interpretable change at top-K → does not weaken AlphaMissense alone, but flag for caution

When you have a saturation question (e.g. "score all 19 substitutions at residue 600 to find the most disruptive"): use ESM_score_variant_sae_batch — 1 Forge call for the reference + 1 per variant, instead of 2 per variant.

When the region is what matters (e.g. "what's the SAE signature of the kinase activation loop, residues 754-771"): use ESM_get_region_sae_features then ESM_describe_sae_feature on the top hits.

Requires: ESM_API_KEY env var (free non-commercial token at https://forge.evolutionaryscale.ai) and pip install 'esm @ git+https://github.com/evolutionaryscale/esm@ee891c52' (SAE support is on an unmerged feature branch — PyPI esm 3.2.x does NOT include SAEConfig). License: EvolutionaryScale Cambrian Inference License — non-commercial use only.

Phase 4.5: Expression Context

Tools: CELLxGENE_get_expression_data, CELLxGENE_get_cell_metadata, GTEx_get_median_gene_expression

Confirms gene expression in disease-relevant tissues. Supports PP4 if highly restricted; challenges classification if not expressed in affected tissue.

Phase 5: Literature Evidence

Tools: PubMed_search_articles, EuropePMC_search_articles, BioRxiv_list_recent_preprints, MedRxiv_get_preprint, openalex_search_works, SemanticScholar_search_papers

Always flag preprints as NOT peer-reviewed.

Phase 6: ACMG Classification

Apply all relevant evidence codes (PVS1, PS1, PS3, PM1, PM2, PM5, PP3, PP5 for pathogenic; BA1, BS1, BS3, BP4, BP7 for benign). See ACMG_CLASSIFICATION.md for the complete algorithm.

Gene-Specific Population Frequency Thresholds

BS1 (allele frequency too high for disorder) requires gene-specific calibration, not a universal cutoff:

• High-penetrance genes (BRCA1, TP53): BS1 threshold ~0.0001
• Moderate-penetrance genes (PALB2, ATM, CHEK2): BS1 threshold ~0.001
• Low-penetrance/common disease genes: BS1 threshold higher, depends on disease prevalence
• Formula: BS1 threshold = (disease prevalence × max allelic contribution × max genetic contribution) / penetrance
• When in doubt, compare the variant's AF to the highest AF of any known pathogenic variant in the same gene — if it exceeds that, BS1 is likely applicable.

Handling Conflicting Evidence: Functional vs Epidemiological

This is one of the most challenging scenarios in variant interpretation. When a biochemical assay shows damage but population/epidemiological data shows no disease association:

1. Epidemiological data generally trumps in-vitro assays for clinical classification. A variant found at ~0.1% frequency with no disease association in 40K+ cases is unlikely to be clinically significant, even if it reduces protein function in a tube. 2. Apply PS3/BS3 carefully: ClinGen's SVI recommends that PS3 (functional evidence for pathogenicity) requires the assay to be validated against known pathogenic AND known benign controls. A single biochemical study without such validation is PS3_Supporting at best. 3. Hypomorphic variants: Some variants genuinely reduce protein function (detectable in sensitive assays) but not enough to cause disease. This is biologically real and does not make them pathogenic. 4. Document the conflict explicitly in the report. State: "Biochemical assay X shows [result], but case-control study Y with N cases found no significant disease association. Per ACMG guidelines, the epidemiological evidence is weighted more heavily for clinical classification."

Bayesian ACMG Point System (Tavtigian et al. 2018)

Modern clinical labs use a point-based system instead of the original rule-counting approach:

Classification by total points:

• Pathogenic: >= 10 points
• Likely Pathogenic: 6-9 points
• VUS: -5 to 5 points
• Likely Benign: -6 to -9 points
• Benign: <= -10 points

This system handles conflicting evidence naturally — a variant with PS3 (+4) and BS1 (-4) and BP4 (-1) nets -1, which is VUS. The original rule-based approach struggles with this scenario.

Computational procedure: ACMG Bayesian classification

# Automated ACMG point calculation
# Input: dict of evidence codes with their applied strength

def classify_acmg(evidence: dict) -> dict:
    """
    Classify a variant using the Bayesian ACMG point system.

    Args:
        evidence: dict mapping ACMG codes to strength levels.
            Pathogenic codes: 'very_strong', 'strong', 'moderate', 'supporting'
            Benign codes: 'stand_alone', 'strong', 'supporting'

    Example:
        evidence = {
            'BS1': 'strong',       # AF too high
            'BS3': 'supporting',   # Epidemiological evidence against pathogenicity
            'BP6': 'supporting',   # ClinVar benign consensus
            'PP3': 'supporting',   # Computational predictors say damaging
        }
    """
    pathogenic_points = {
        'very_strong': 8, 'strong': 4, 'moderate': 2, 'supporting': 1
    }
    benign_points = {
        'stand_alone': -8, 'strong': -4, 'supporting': -1
    }

    total = 0
    details = []
    for code, strength in evidence.items():
        if code.startswith(('PVS', 'PS', 'PM', 'PP')):
            pts = pathogenic_points.get(strength, 0)
        elif code.startswith(('BA', 'BS', 'BP')):
            pts = benign_points.get(strength, 0)
        else:
            pts = 0
        total += pts
        details.append(f"{code} ({strength}): {pts:+d}")

    if total >= 10:
        classification = "Pathogenic"
    elif 6 <= total <= 9:
        classification = "Likely Pathogenic"
    elif -5 <= total <= 5:
        classification = "VUS"
    elif -9 <= total <= -6:
        classification = "Likely Benign"
    else:
        classification = "Benign"

    return {
        'classification': classification,
        'total_points': total,
        'evidence_breakdown': details
    }

# Example: PALB2 c.2816T>G (from test case)
result = classify_acmg({
    'BS1': 'strong',       # gnomAD AF 0.00105 exceeds threshold
    'BS3': 'supporting',   # Case-control study shows no association
    'BP6': 'supporting',   # ClinVar 13 submitters say benign/likely benign
})
# Output: Likely Benign, total_points=-6, evidence: BS1(strong):-4, BS3(supporting):-1, BP6(supporting):-1

Use this procedure after collecting all evidence from Phases 1-5 to compute the final classification.

Gene-Specific VCEP Criteria

ClinGen Variant Curation Expert Panels (VCEPs) publish gene-specific ACMG modifications. Before classifying, check if a VCEP exists:

• ClinGen_search_gene_validity(gene="<gene_symbol>") — if validity is "Definitive" or "Strong", a VCEP likely exists
• Common VCEPs: BRCA1/2 (Enigma), TP53, PTEN, CDH1, PALB2, RASopathies, Lynch syndrome genes
• VCEP criteria override generic ACMG criteria (e.g., PALB2 VCEP has specific PM1 hotspot regions)

Predictor Weighting

Not all computational predictors are equal. For missense variants:

• REVEL (AUC ~0.95) — best single meta-predictor; weight highest
• AlphaMissense (AUC ~0.94) — strong, structure-aware
• CADD (AUC ~0.85) — good for all variant types, but less specific for missense
• SIFT/PolyPhen (AUC ~0.80) — legacy tools; useful for consensus but not individually decisive

When predictors disagree: if REVEL says tolerated but SIFT/PolyPhen say damaging, lean toward REVEL. If REVEL is unavailable, require 3+ concordant predictions for PP3/BP4.

Tool Failure Fallbacks

If a primary tool fails, use these alternatives:

• ClinVar_search_variants returns 0 results: Use MyVariant_query_variants with rsID or HGVS — the clinvar field in MyVariant is more reliable for variant lookup than NCBI Entrez search
• gnomad_search_variants fails: Use EnsemblVEP_annotate_hgvs which includes gnomAD frequency via colocated variants
• CADD_get_variant_score fails: CADD PHRED is also available in the dbnsfp block from MyVariant
• AlphaFold prediction unavailable (large proteins >2700aa): Use PDBe_get_uniprot_mappings for experimental structures

---

Special Scenarios

Novel Missense VUS: Check PM5 (other pathogenic at same residue), get AlphaFold2 structure, apply PM1/PP3 as appropriate.

Truncating Variant: Check LOF mechanism, NMD escape, alternative isoforms, ClinGen LOF curation. Apply PVS1 at appropriate strength.

Splice Variant: Run SpliceAI, assess canonical splice distance, in-frame skipping potential. Apply PP3/BP7 based on scores.

---

Output Structure

# Variant Interpretation Report: {GENE} {VARIANT}
## Executive Summary
## 1. Variant Identity
## 2. Population Data
## 3. Clinical Database Evidence
## 4. Computational Predictions
## 5. Structural Analysis
## 6. Literature Evidence
## 7. ACMG Classification
## 8. Clinical Recommendations
## 9. Limitations & Uncertainties
## Data Sources

File naming: {GENE}_{VARIANT}_interpretation_report.md

---

Clinical Recommendations

Pathogenic/Likely Pathogenic: Enhanced screening, risk-reducing options, drug dosing adjustment, reproductive counseling, family cascade screening.

VUS: Do not use for medical decisions. Reinterpret in 1-2 years. Pursue functional studies and segregation data.

Benign/Likely Benign: Not expected to cause disease. No cascade testing needed.

---

Quantified Minimums

---

Cross-Skill References

For amino acid properties at variant position, run: python3 skills/tooluniverse-sequence-analysis/scripts/amino_acids.py --type amino_acid --code X

---

References

• ACMG_CLASSIFICATION.md - Evidence codes, classification algorithm, prediction thresholds, structural/regulatory impact tables
• CODE_PATTERNS.md - Reusable code patterns for each workflow phase
• CHECKLIST.md - Pre-delivery verification
• EXAMPLES.md - Sample interpretations
• TOOLS_REFERENCE.md - Tool parameters and fallbacks

ACMG Classification Reference

Evidence Codes

Pathogenic Evidence

Benign Evidence

Classification Algorithm

Classification Confidence

ClinVar Classification Map

gnomAD Frequency Thresholds (Rare Disease)

COSMIC Somatic Evidence

DisGeNET Score Interpretation

ClinGen Validity Levels (for ACMG PM1/PP4)

ClinGen Dosage Sensitivity Scores (for CNV interpretation)

Structural Impact Categories

Structural Impact Confidence (AlphaFold pLDDT)

Prediction Thresholds

PP3/BP4 Application Notes

• PP3: Multiple concordant damaging predictions (AlphaMissense + CADD + EVE agreement = strong PP3)
• BP4: Multiple concordant benign predictions
• Note: AlphaMissense alone achieves ~90% accuracy on ClinVar pathogenic variants

SpliceAI Thresholds

Literature Evidence Weights

Regulatory Impact Categories

PVS1 Application for Truncating Variants

Clinical Variant Interpreter Checklist

Pre-delivery verification checklist for variant interpretation reports.

Report Quality Checklist

Structure & Format

• [ ] Report file created: {GENE}_{VARIANT}_interpretation_report.md
• [ ] All 9 main sections present
• [ ] Executive summary completed (not [Interpreting...])
• [ ] Data sources section populated

Phase 1: Variant Identity

• [ ] Gene symbol identified
• [ ] HGVS c. notation provided
• [ ] HGVS p. notation (if applicable)
• [ ] Transcript ID (MANE Select preferred)
• [ ] Consequence type identified
• [ ] Exon/intron location stated
• [ ] Amino acid change (for missense)

Phase 2: Population Data

• [ ] gnomAD queried
• [ ] Overall allele frequency reported
• [ ] ≥3 ancestry-specific frequencies
• [ ] Homozygote count reported
• [ ] Hemizygote count (X-linked genes)
• [ ] Frequency interpreted vs. disease prevalence

Phase 3: Clinical Database Evidence

• [ ] ClinVar searched
• [ ] Classification reported (or "Not in ClinVar")
• [ ] Review status noted (gold stars)
• [ ] Number of submissions documented
• [ ] Conflicting interpretations noted (if any)
• [ ] OMIM gene-disease associations checked
• [ ] ClinGen gene validity (if available)

Phase 4: Computational Predictions

• [ ] ≥3 predictors reported
• [ ] SIFT score and interpretation
• [ ] PolyPhen-2 score and interpretation
• [ ] CADD score (raw and phred)
• [ ] Concordance assessed (agree/disagree)
• [ ] PP3 or BP4 applied appropriately

Phase 5: Structural Analysis (for missense)

• [ ] Protein structure source identified (PDB/AlphaFold)
• [ ] pLDDT at variant position (if AlphaFold)
• [ ] Residue location (buried/surface)
• [ ] Secondary structure context
• [ ] Domain/functional site proximity
• [ ] PM1 consideration documented

Phase 6: Literature Evidence

• [ ] PubMed searched with ≥2 strategies
• [ ] Functional studies documented (or "None found")
• [ ] Case reports documented
• [ ] PS3 consideration documented
• [ ] PP1 (segregation) documented if available

Phase 7: ACMG Classification

• [ ] All evidence codes explicitly listed
• [ ] Each code has strength modifier
• [ ] Code justification provided
• [ ] Classification calculated correctly
• [ ] Classification stated in executive summary

Phase 8: Clinical Recommendations

• [ ] Recommendations appropriate to classification
• [ ] VUS: No medical decisions
• [ ] Pathogenic: Specific follow-up
• [ ] Family screening addressed

Phase 9: Limitations

• [ ] Missing data acknowledged
• [ ] Conflicting evidence noted
• [ ] Uncertainty quantified

---

Citation Requirements

Every Evidence Statement Must Include

• [ ] Source database/tool in backticks
• [ ] Specific identifier (ClinVar ID, PMID)
• [ ] Date of access (for changing databases)

Format Examples

*Source: ClinVar VCV000012345, reviewed 4-star, accessed 2026-02-04*
*Source: gnomAD v4.0, overall AF=0.00001, accessed 2026-02-04*
*Source: PMID 12345678 - functional study showed loss of activity*
*Source: AlphaFold DB via `alphafold_get_prediction`, pLDDT=92 at position*

---

ACMG Code Verification

For Each Code Applied

• [ ] Code abbreviation correct (PVS1, PM2, PP3, etc.)
• [ ] Strength appropriate (VeryStrong/Strong/Moderate/Supporting)
• [ ] Evidence clearly supports application
• [ ] Not double-counted

Common Errors to Avoid

• [ ] PM2 without checking gnomAD
• [ ] PP3 without multiple concordant predictions
• [ ] PVS1 for non-null variants
• [ ] PS3 without true functional evidence
• [ ] Applying same evidence to multiple codes

---

Classification Calculation

Verify Math

Classification Cross-Check

• [ ] Evidence codes align with final classification
• [ ] No conflicting codes ignored
• [ ] Classification matches standard algorithms

---

Evidence Grading

All Classifications Must Have

• [ ] Classification tier: ★★★, ★★☆, ★☆☆, or VUS
• [ ] Evidence strength description
• [ ] Key supporting evidence highlighted

Tier Definitions

---

Quantified Minimums

---

Structural Analysis Quality (for Missense)

Must Include

• [ ] Structure source (PDB ID or "AlphaFold predicted")
• [ ] pLDDT at position (if AlphaFold)
• [ ] Residue depth/accessibility
• [ ] Structural consequence prediction

Quality Thresholds

---

Special Scenario Checks

Truncating Variants

• [ ] NMD prediction assessed
• [ ] LOF mechanism confirmed for gene
• [ ] PVS1 strength correctly applied
• [ ] Last exon exception considered

Splice Variants

• [ ] Canonical splice site distance
• [ ] SpliceAI scores (if available)
• [ ] In-frame skip assessment
• [ ] PVS1 modified if warranted

X-linked Genes

• [ ] Sex of individual considered
• [ ] Hemizygote frequency used appropriately
• [ ] Penetrance in females addressed

---

Output Files

Required

• [ ] {GENE}_{VARIANT}_interpretation_report.md - Main report

Optional Data Export

• [ ] {GENE}_{VARIANT}_evidence_table.csv - Structured evidence
• [ ] {GENE}_{VARIANT}_acmg_codes.csv - Applied codes

---

Final Review

Before Delivery

• [ ] No [Interpreting...] placeholders remaining
• [ ] All tables properly formatted
• [ ] Executive summary synthesizes findings
• [ ] Classification stated prominently
• [ ] Clinical recommendations actionable
• [ ] Limitations clearly stated

Common Issues to Avoid

• [ ] Missing gnomAD frequencies
• [ ] Classification without evidence codes
• [ ] Recommendations not matching classification
• [ ] Missing literature search
• [ ] Structure analysis skipped for VUS missense
• [ ] ACMG codes without justification

---

ClinVar-Specific Checks

When Variant in ClinVar

• [ ] VCV ID documented
• [ ] Review status (stars) noted
• [ ] Number of submitters
• [ ] Date of last evaluation
• [ ] Concordance with our assessment

When Variant NOT in ClinVar

• [ ] Explicitly state "Not in ClinVar as of {date}"
• [ ] Consider novel variant workflow
• [ ] Emphasize structural analysis

---

Tool Verification Checklist

Before Report Generation

• [ ] ClinVar_search_variants returns results or "not found"
• [ ] gnomad_search_variants frequency values valid
• [ ] MyVariant_query_variants predictions populated
• [ ] Structure available (PDB or AlphaFold)

NVIDIA NIM Availability

• [ ] Check if structural analysis needed
• [ ] Confirm NVIDIA_API_KEY if using NvidiaNIM_alphafold2
• [ ] Document if fallback used

---

Recommendations Matrix

Match Recommendations to Classification

---

Report Completeness Score

Calculate before delivery:

Minimum passing score: 85/100

Code Patterns for Variant Interpretation

Reusable code patterns for each phase of the variant interpretation workflow.

---

Phase 2: Clinical Database Code Patterns

2.1 COSMIC Somatic Context

def get_somatic_context(tu, gene_symbol, variant_aa):
    """Get somatic mutation context from COSMIC."""

    # Search for specific mutation
    cosmic = tu.tools.COSMIC_search_mutations(
        operation="search",
        terms=f"{gene_symbol} {variant_aa}",
        max_results=20,
        genome_build=38
    )

    # Get all gene mutations for context
    gene_mutations = tu.tools.COSMIC_get_mutations_by_gene(
        operation="get_by_gene",
        gene=gene_symbol,
        max_results=100
    )

    # Determine if it's a hotspot
    mutation_counts = Counter(m['MutationAA'] for m in gene_mutations.get('results', []))
    is_hotspot = variant_aa in [m[0] for m in mutation_counts.most_common(10)]

    return {
        'cosmic_hits': cosmic.get('results', []),
        'is_somatic_hotspot': is_hotspot,
        'cancer_types': [m['PrimarySite'] for m in cosmic.get('results', [])],
        'total_cosmic_count': cosmic.get('total_count', 0)
    }

2.2 OMIM Gene-Disease Context

def get_omim_context(tu, gene_symbol):
    """Get OMIM gene-disease associations."""

    search = tu.tools.OMIM_search(
        operation="search",
        query=gene_symbol,
        limit=5
    )

    omim_data = []
    for entry in search.get('data', {}).get('entries', []):
        mim = entry.get('mimNumber')

        details = tu.tools.OMIM_get_entry(
            operation="get_entry",
            mim_number=str(mim)
        )

        synopsis = tu.tools.OMIM_get_clinical_synopsis(
            operation="get_clinical_synopsis",
            mim_number=str(mim)
        )

        omim_data.append({
            'mim_number': mim,
            'title': details.get('data', {}).get('titles', {}),
            'inheritance': synopsis.get('data', {}).get('inheritance'),
            'clinical_features': synopsis.get('data', {})
        })

    return omim_data

2.3 DisGeNET Gene-Disease Evidence

def get_disgenet_context(tu, gene_symbol, variant_rsid=None):
    """Get gene-disease associations from DisGeNET."""

    gda = tu.tools.DisGeNET_search_gene(
        operation="search_gene",
        gene=gene_symbol,
        limit=20
    )

    vda = None
    if variant_rsid:
        vda = tu.tools.DisGeNET_get_vda(
            operation="get_vda",
            variant=variant_rsid,
            limit=20
        )

    return {
        'gene_associations': gda.get('data', {}).get('associations', []),
        'variant_associations': vda.get('data', {}).get('associations', []) if vda else []
    }

2.4 ClinGen Gene Validity & Dosage Sensitivity

def get_clingen_evidence(tu, gene_symbol):
    """
    Get ClinGen gene validity and dosage sensitivity data.
    CRITICAL for ACMG classification - establishes gene-disease validity.
    """

    validity = tu.tools.ClinGen_search_gene_validity(gene=gene_symbol)

    validity_data = []
    if validity.get('data'):
        for entry in validity.get('data', []):
            validity_data.append({
                'disease': entry.get('Disease Label'),
                'classification': entry.get('Classification'),
                'inheritance': entry.get('Inheritance'),
                'mondo_id': entry.get('Disease ID (MONDO)')
            })

    dosage = tu.tools.ClinGen_search_dosage_sensitivity(gene=gene_symbol)

    dosage_data = {}
    if dosage.get('data'):
        for entry in dosage.get('data', []):
            dosage_data = {
                'haploinsufficiency_score': entry.get('Haploinsufficiency Score'),
                'triplosensitivity_score': entry.get('Triplosensitivity Score'),
                'disease': entry.get('Disease')
            }
            break

    actionability = tu.tools.ClinGen_search_actionability(gene=gene_symbol)

    return {
        'gene_validity': validity_data,
        'dosage_sensitivity': dosage_data,
        'actionability': actionability.get('data', {}),
        'has_definitive_validity': any(v['classification'] == 'Definitive' for v in validity_data),
        'is_haploinsufficient': dosage_data.get('haploinsufficiency_score') == '3'
    }

2.5 SpliceAI Splice Variant Prediction

def get_spliceai_prediction(tu, chrom, pos, ref, alt, genome="38"):
    """
    Get SpliceAI splice effect predictions.

    Delta scores: DS_AG (Acceptor gain), DS_AL (Acceptor loss),
                  DS_DG (Donor gain), DS_DL (Donor loss)
    """

    variant = f"chr{chrom}-{pos}-{ref}-{alt}"

    result = tu.tools.SpliceAI_predict_splice(
        variant=variant,
        genome=genome
    )

    if result.get('data'):
        max_score = result['data'].get('max_delta_score', 0)
        interpretation = result['data'].get('interpretation', '')

        if max_score >= 0.8:
            acmg = 'PP3 (strong) - high splice impact'
        elif max_score >= 0.5:
            acmg = 'PP3 (supporting) - moderate splice impact'
        elif max_score >= 0.2:
            acmg = 'PP3 (weak) - possible splice impact'
        else:
            acmg = 'BP7 (if synonymous) - splice benign'

        return {
            'max_delta_score': max_score,
            'interpretation': interpretation,
            'acmg_support': acmg,
            'scores': result['data'].get('scores', [])
        }
    return None

def quick_splice_check(tu, variant, genome="38"):
    """Quick triage using max delta score only."""
    result = tu.tools.SpliceAI_get_max_delta(variant=variant, genome=genome)
    return result.get('data', {})

When to Use SpliceAI:

• Intronic variants near splice sites (+/-50bp)
• Synonymous variants (may still affect splicing)
• Exonic variants near splice junctions
• Variants creating cryptic splice sites

---

Phase 2.5: Regulatory Context (Non-Coding Variants)

def assess_regulatory_impact(tu, variant_position, gene_symbol):
    """Assess regulatory impact of non-coding variant."""

    tf_binding = tu.tools.ChIPAtlas_enrichment_analysis(
        gene=gene_symbol,
        cell_type="all"
    )

    peaks = tu.tools.ChIPAtlas_get_peak_data(
        gene=gene_symbol,
        experiment_type="TF"
    )

    encode_data = tu.tools.ENCODE_search_experiments(
        assay_title="ATAC-seq",
        biosample="all"
    )

    binding_disrupted = check_motif_disruption(variant_position, peaks)

    return {
        'tf_binding': tf_binding,
        'regulatory_peaks': peaks,
        'encode_annotations': encode_data,
        'likely_regulatory': binding_disrupted
    }

---

Phase 3: Computational Predictions

3.1 CADD Scoring

def get_cadd_score(tu, chrom, pos, ref, alt):
    """Get CADD deleteriousness score for a variant."""

    result = tu.tools.CADD_get_variant_score(
        chrom=str(chrom),
        pos=pos,
        ref=ref,
        alt=alt,
        version="GRCh38-v1.7"
    )

    if result.get('status') == 'success':
        phred = result['data'].get('phred_score')
        return {
            'score': phred,
            'interpretation': result['data'].get('interpretation'),
            'acmg_support': 'PP3' if phred >= 20 else ('BP4' if phred < 15 else 'neutral')
        }
    return None

3.2 AlphaMissense Pathogenicity

def get_alphamissense_score(tu, uniprot_id, variant):
    """
    Get AlphaMissense pathogenicity score.
    variant format: 'R123H' or 'p.R123H'

    Thresholds: Pathogenic >0.564, Ambiguous 0.34-0.564, Benign <0.34
    """

    result = tu.tools.AlphaMissense_get_variant_score(
        uniprot_id=uniprot_id,
        variant=variant
    )

    if result.get('status') == 'success' and result.get('data'):
        score = result['data'].get('pathogenicity_score')
        classification = result['data'].get('classification')

        if classification == 'pathogenic':
            acmg = 'PP3 (strong)'
        elif classification == 'benign':
            acmg = 'BP4 (strong)'
        else:
            acmg = 'neutral'

        return {
            'score': score,
            'classification': classification,
            'acmg_support': acmg
        }
    return None

3.3 EVE Evolutionary Prediction

def get_eve_score(tu, chrom, pos, ref, alt):
    """Get EVE evolutionary pathogenicity score. Threshold: >0.5 = likely pathogenic."""

    result = tu.tools.EVE_get_variant_score(
        chrom=str(chrom), pos=pos, ref=ref, alt=alt
    )

    if result.get('status') == 'success':
        eve_scores = result['data'].get('eve_scores', [])
        if eve_scores:
            best_score = eve_scores[0]
            return {
                'score': best_score.get('eve_score'),
                'classification': best_score.get('classification'),
                'gene': best_score.get('gene_symbol'),
                'acmg_support': 'PP3' if best_score.get('eve_score', 0) > 0.5 else 'BP4'
            }
    return None

3.4 Integrated Prediction Strategy

def comprehensive_pathogenicity_assessment(tu, variant_info):
    """Combine all prediction tools for robust classification."""
    chrom = variant_info['chrom']
    pos = variant_info['pos']
    ref = variant_info['ref']
    alt = variant_info['alt']
    uniprot_id = variant_info.get('uniprot_id')
    aa_change = variant_info.get('aa_change')

    predictions = {}

    cadd = get_cadd_score(tu, chrom, pos, ref, alt)
    if cadd:
        predictions['cadd'] = cadd

    if uniprot_id and aa_change:
        am = get_alphamissense_score(tu, uniprot_id, aa_change)
        if am:
            predictions['alphamissense'] = am

    eve = get_eve_score(tu, chrom, pos, ref, alt)
    if eve:
        predictions['eve'] = eve

    damaging_count = sum(1 for p in predictions.values()
                         if 'PP3' in p.get('acmg_support', ''))
    benign_count = sum(1 for p in predictions.values()
                       if 'BP4' in p.get('acmg_support', ''))

    if damaging_count >= 2 and benign_count == 0:
        consensus = 'likely_damaging'
        acmg = 'PP3 (multiple predictors concordant)'
    elif benign_count >= 2 and damaging_count == 0:
        consensus = 'likely_benign'
        acmg = 'BP4 (multiple predictors concordant)'
    else:
        consensus = 'uncertain'
        acmg = 'neutral (discordant predictions)'

    return {
        'predictions': predictions,
        'consensus': consensus,
        'acmg_recommendation': acmg
    }

---

Phase 4: Structural Analysis (AlphaFold2)

# 1. Get protein sequence
protein_seq = tu.tools.UniProt_get_sequence_by_accession(accession=uniprot_id)

# 2. Get/predict structure
try:
    pdb_hits = tu.tools.PDBe_get_uniprot_mappings(uniprot_id=uniprot_id)
    structure = tu.tools.PDB_get_structure(pdb_id=pdb_hits[0]['pdb_id'])
except:
    structure = tu.tools.NvidiaNIM_alphafold2(
        sequence=protein_seq['sequence'],
        algorithm="mmseqs2"
    )

# 3. Analyze variant position
# - Extract pLDDT at residue position
# - Calculate solvent accessibility
# - Check for nearby functional sites

Structural Features to Report:

• pLDDT at variant position
• Secondary structure (helix/sheet/coil)
• Solvent accessibility (buried/exposed)
• Distance to active site (if applicable)
• Interactions disrupted (H-bonds, salt bridges)

---

Phase 4.5: Expression Context

def validate_expression_context(tu, gene_symbol, phenotype_tissues):
    """Validate gene is expressed in phenotype-relevant tissues."""

    sc_expression = tu.tools.CELLxGENE_get_expression_data(
        gene=gene_symbol,
        tissue=phenotype_tissues[0] if phenotype_tissues else "all"
    )

    gtex = tu.tools.GTEx_get_median_gene_expression(gene=gene_symbol)

    relevant_expression = {
        tissue: gtex.get(tissue, 0)
        for tissue in phenotype_tissues
    }

    return {
        'single_cell': sc_expression,
        'gtex': relevant_expression,
        'expressed_in_phenotype_tissue': any(v > 1 for v in relevant_expression.values())
    }

---

Phase 5: Literature Search

def comprehensive_literature_search(tu, gene, variant, phenotype):
    """Search across all literature sources."""

    pubmed = tu.tools.PubMed_search_articles(
        query=f'"{gene}" AND ("{variant}" OR functional)',
        max_results=30
    )

    biorxiv = tu.tools.BioRxiv_list_recent_preprints(
        query=f"{gene} {phenotype}",
        limit=10
    )

    medrxiv = tu.tools.MedRxiv_get_preprint(
        query=f"{gene} variant {phenotype}",
        limit=10
    )

    key_papers = pubmed[:5]
    for paper in key_papers:
        citations = tu.tools.openalex_search_works(
            query=paper['title'],
            limit=1
        )
        paper['citation_count'] = citations[0].get('cited_by_count', 0) if citations else 0

    return {
        'pubmed': pubmed,
        'preprints': biorxiv + medrxiv,
        'key_papers_with_citations': key_papers
    }

Search Query Templates:

# Gene + variant specific
"{GENE} AND ({HGVS_p} OR {AA_change})"

# Functional studies
"{GENE} AND (functional OR functional study OR mutagenesis)"

# Clinical reports
"{GENE} AND (case report OR patient) AND {phenotype}"

# Preprint-specific
"{GENE} genetics 2024" (for recent preprints)

Warning: Always flag preprints as NOT peer-reviewed in reports.

Clinical Variant Interpreter - Examples

Example 1: Pathogenic BRCA1 Variant

User Query

"Interpret the clinical significance of BRCA1 c.5266dupC"

Expected Output

# Variant Interpretation Report: BRCA1 c.5266dupC

**Generated**: 2026-02-04 | **Analyst**: ToolUniverse Variant Interpreter

---

## Executive Summary

| Field | Value |
|-------|-------|
| **Variant** | NM_007294.4:c.5266dupC (p.Gln1756ProfsTer74) |
| **Gene** | BRCA1 |
| **Classification** | **PATHOGENIC** ★★★ |
| **Evidence Strength** | Very Strong (multiple independent lines) |
| **Key Finding** | Frameshift variant in established LOF gene; well-documented pathogenic variant |

**Clinical Action Required**: Hereditary breast/ovarian cancer risk management indicated.

---

## 1. Variant Identity

| Property | Value |
|----------|-------|
| Gene | BRCA1 |
| Chromosome | 17 |
| Position | 17:43057051 (GRCh38) |
| Transcript | NM_007294.4 (MANE Select) |
| cDNA change | c.5266dupC |
| Protein change | p.Gln1756ProfsTer74 |
| Consequence | Frameshift |
| Exon | 20 of 23 |

**Note**: Also known as 5382insC (legacy notation)

*Source: `EnsemblVar_get_variant_consequences`, `MyVariant_query_variants`*

---

## 2. Population Data

### gnomAD v4.0 Frequencies

| Population | Allele Frequency | Allele Count |
|------------|------------------|--------------|
| **Overall** | 0.00006 | 45 |
| European (Non-Finnish) | 0.00012 | 38 |
| Ashkenazi Jewish | 0.011 | 89 |
| African/African American | 0.00001 | 2 |
| Latino/Admixed American | 0.00002 | 3 |
| East Asian | 0 | 0 |
| South Asian | 0 | 0 |

**Homozygotes**: 0
**Hemizygotes**: N/A (autosomal gene)

**Interpretation**: Elevated in Ashkenazi Jewish population (founder variant). Overall frequency consistent with disease prevalence when accounting for reduced penetrance and carrier frequency.

*Source: `gnomad_search_variants`, accessed 2026-02-04*

---

## 3. Clinical Database Evidence

### ClinVar

| Property | Value |
|----------|-------|
| VCV ID | VCV000017661 |
| Classification | Pathogenic |
| Review Status | ★★★★ (Expert panel reviewed) |
| Submitters | 47 |
| Last Evaluated | 2024-12-15 |
| Conditions | Hereditary breast/ovarian cancer syndrome |

**Expert Panel**: ENIGMA consortium - criteria met for Pathogenic

### OMIM

| Gene-Disease | Inheritance | MIM# |
|--------------|-------------|------|
| Breast-ovarian cancer, familial 1 | AD | 604370 |
| Fanconi anemia, complementation group S | AR | 617883 |

### ClinGen

| Assessment | Status |
|------------|--------|
| Gene-Disease Validity | DEFINITIVE for HBOC |
| Dosage Sensitivity | Haploinsufficient |
| LOF Mechanism | Established |

*Sources: ClinVar VCV000017661, OMIM #113705, ClinGen*

---

## 4. Computational Predictions

**Note**: Computational predictors not applicable to frameshift variants (by definition, these are loss-of-function).

| Predictor | Score | Not Applicable Reason |
|-----------|-------|----------------------|
| SIFT | N/A | Frameshift |
| PolyPhen-2 | N/A | Frameshift |
| CADD | 35 (Phred) | High deleteriousness |

**NMD Prediction**: Predicted to undergo nonsense-mediated decay (PTC at codon 1829, well before last exon junction)

*Source: `MyVariant_query_variants`*

---

## 5. Structural Analysis

**Structural impact assessment not required** for truncating variants where LOF is established mechanism.

**Protein Context**:
- Truncation at codon 1756 (of 1863)
- Removes C-terminal BRCT domains
- BRCT domains essential for DNA repair function
- Even if translated, protein would be non-functional

*Source: UniProt P38398, InterPro domain annotations*

---

## 6. Literature Evidence

### Functional Studies

| Study | Finding | PMID |
|-------|---------|------|
| Carvalho et al., 2007 | LOF in HDR assay | 17308087 |
| Moghadasi et al., 2018 | Absent protein | 29446198 |
| Multiple studies | Segregates with cancer | Multiple |

### Clinical Reports

- Founder variant in Ashkenazi Jewish population
- Extensive clinical documentation (>500 families)
- Co-segregation in >100 affected individuals

**PS3 Evidence**: Strong - multiple functional studies demonstrating loss of DNA repair function

*Source: `PubMed_search_articles`*

---

## 7. ACMG Classification

### Evidence Codes Applied

| Code | Strength | Rationale |
|------|----------|-----------|
| **PVS1** | Very Strong | Null variant (frameshift) in gene where LOF is established mechanism |
| **PS4** | Strong | >5 affected individuals documented |
| **PM2** | Supporting | Absent/extremely rare in general population (outside founders) |
| **PP5** | Supporting | Expert panel (ENIGMA) reports pathogenic |

### Evidence Summary

| Pathogenic | Benign |
|------------|--------|
| 1 Very Strong (PVS1) | None |
| 1 Strong (PS4) | |
| 2 Supporting (PM2, PP5) | |

### Classification: **PATHOGENIC**

**Rationale**: Meets ACMG criteria for Pathogenic (1 Very Strong + 1 Strong + 2 Supporting). Well-established pathogenic variant with multiple lines of independent evidence.

**Confidence**: ★★★ (High) - Expert panel reviewed, extensive clinical data

---

## 8. Clinical Recommendations

### For Affected Individual

| Category | Recommendation |
|----------|----------------|
| **Confirmation** | No re-testing needed; well-documented variant |
| **Cancer screening** | NCCN high-risk breast/ovarian protocols |
| **Risk-reducing options** | Discuss prophylactic mastectomy, salpingo-oophorectomy |
| **Medical management** | Consider PARP inhibitor eligibility if cancer |
| **Genetic counseling** | Discuss implications, reproductive options |

### For Family Members

| Relationship | Action |
|--------------|--------|
| First-degree relatives | Offer cascade testing |
| At-risk children | Consider testing at 18-25 years |
| Partners | Reproductive counseling if planning children |

### Testing Recommendations

- Cascade genetic testing for first-degree relatives
- Consider panel testing for other cancer genes if clinically indicated
- Tumor testing for PARP inhibitor eligibility if cancer present

---

## 9. Limitations & Uncertainties

| Factor | Status |
|--------|--------|
| Classification certainty | High - well-documented |
| Penetrance | Incomplete (~70% breast, ~40% ovarian by age 70) |
| Modifier genes | May influence risk |
| Environmental factors | May modify risk |

**No significant limitations** - This is a well-characterized pathogenic variant with extensive clinical documentation.

---

## Data Sources

| Tool | Purpose | Result |
|------|---------|--------|
| `ClinVar_search_variants` | Classification lookup | VCV000017661 |
| `gnomad_search_variants` | Population frequency | AF=0.00006 |
| `MyVariant_query_variants` | Aggregated annotations | Complete |
| `PubMed_search_articles` | Literature evidence | 47 relevant papers |
| `OMIM_search` | Gene-disease | HBOC confirmed |

---

Example 2: VUS Missense Requiring Structural Analysis

User Query

"Interpret TP53 p.R175C - it's reported as a VUS"

Expected Output

# Variant Interpretation Report: TP53 p.R175C

**Generated**: 2026-02-04 | **Analyst**: ToolUniverse Variant Interpreter

---

## Executive Summary

| Field | Value |
|-------|-------|
| **Variant** | NM_000546.6:c.523C>T (p.Arg175Cys) |
| **Gene** | TP53 |
| **Classification** | **LIKELY PATHOGENIC** ★★☆ |
| **Evidence Strength** | Moderate (structural + computational + clinical) |
| **Key Finding** | Hotspot residue in DNA-binding domain; R175 is known pathogenic position (R175H documented) |

**Reclassified from VUS based on structural analysis and PM5 evidence.**

---

## 1. Variant Identity

| Property | Value |
|----------|-------|
| Gene | TP53 |
| Chromosome | 17 |
| Position | 17:7674220 (GRCh38) |
| Transcript | NM_000546.6 (MANE Select) |
| cDNA change | c.523C>T |
| Protein change | p.Arg175Cys |
| Consequence | Missense |
| Exon | 5 of 11 |

*Source: `EnsemblVar_get_variant_consequences`, `MyVariant_query_variants`*

---

## 2. Population Data

### gnomAD v4.0 Frequencies

| Population | Allele Frequency | Allele Count |
|------------|------------------|--------------|
| **Overall** | 0 | 0 |
| European (Non-Finnish) | 0 | 0 |
| African/African American | 0 | 0 |
| Latino/Admixed American | 0 | 0 |
| East Asian | 0 | 0 |
| South Asian | 0 | 0 |

**Homozygotes**: 0

**Interpretation**: Absent from gnomAD (>140,000 individuals). Supports PM2 (absent from controls).

*Source: `gnomad_search_variants`, accessed 2026-02-04*

---

## 3. Clinical Database Evidence

### ClinVar

| Property | Value |
|----------|-------|
| VCV ID | VCV000376642 |
| Classification | VUS (outdated) |
| Review Status | ★★ (criteria provided) |
| Submitters | 3 |
| Conditions | Li-Fraumeni syndrome |

**Note**: ClinVar classification may require update based on structural evidence.

### Critical Context: R175 Position

| Variant at R175 | ClinVar Classification |
|-----------------|------------------------|
| p.R175H | Pathogenic (★★★★) |
| p.R175G | Pathogenic |
| p.R175L | Pathogenic |
| p.R175S | Pathogenic |
| **p.R175C** | VUS (being reclassified) |

**PM5 Applies**: Multiple different missense changes at R175 are pathogenic.

### OMIM

| Gene-Disease | Inheritance |
|--------------|-------------|
| Li-Fraumeni syndrome | AD |
| Various cancers | Somatic |

*Sources: ClinVar, OMIM #191170*

---

## 4. Computational Predictions

| Predictor | Score | Interpretation |
|-----------|-------|----------------|
| SIFT | 0.00 | Damaging |
| PolyPhen-2 | 1.00 | Probably damaging |
| CADD | 29.5 (Phred) | Top 0.1% deleterious |
| REVEL | 0.92 | Pathogenic range |

**Concordance**: 4/4 predictors indicate damaging → **PP3 applies**

*Source: `MyVariant_query_variants`*

---

## 5. Structural Analysis

### Structure Information

| Property | Value |
|----------|-------|
| Structure used | PDB: 2OCJ (1.8 Å) |
| Domain | DNA-binding domain (DBD) |
| Position pLDDT | N/A (experimental structure) |
| Resolution | 1.8 Å |

### R175 Structural Context

| Feature | Assessment |
|---------|------------|
| **Location** | DNA-binding domain core |
| **Solvent accessibility** | Buried (RSA = 5%) |
| **Secondary structure** | Loop L2 (critical for zinc coordination) |
| **Zinc coordination** | 4.2 Å from Zn²⁺ binding site |
| **Conservation** | 100% in vertebrates |

### Structural Impact Analysis

| Factor | Wildtype (Arg) | Mutant (Cys) | Impact |
|--------|----------------|--------------|--------|
| Charge | +1 (basic) | 0 (neutral) | Charge loss in buried position |
| Side chain volume | Large | Small | Potential cavity |
| H-bonding | 5 H-bonds | 1 H-bond | Loss of stabilizing interactions |
| Zinc coordination | Proximal | Potential interference | May affect zinc binding |

### Structural Conclusion

**PM1 applies (moderate)**: 
- R175 is in a critical structural region
- Known mutational hotspot
- Zinc coordination region essential for DNA binding
- Structural analysis supports pathogenicity

*Sources: PDB 2OCJ, structural analysis*

---

## 6. Literature Evidence

### Functional Studies (at R175 position)

| Study | Variant | Finding | PMID |
|-------|---------|---------|------|
| Kato et al., 2003 | R175H | Loss of transactivation | 12826609 |
| Bullock et al., 2000 | R175H | Destabilized, unfolded | 10788335 |
| Joerger et al., 2006 | R175 mutations | Disrupt zinc binding | 16630891 |

### R175C-Specific Evidence

- Limited direct functional data for R175C specifically
- Same position as extensively characterized R175H
- Mechanism (zinc coordination disruption) likely conserved

**PS3**: Not directly applicable (no R175C-specific functional study)
**PS1**: Not applicable (amino acid change differs from R175H)

*Source: `PubMed_search_articles`*

---

## 7. ACMG Classification

### Evidence Codes Applied

| Code | Strength | Rationale |
|------|----------|-----------|
| **PM2** | Moderate | Absent from gnomAD (>140,000 individuals) |
| **PM5** | Moderate | Different missense at same position (R175H) is pathogenic |
| **PM1** | Moderate | Critical DNA-binding domain, zinc coordination region |
| **PP3** | Supporting | All 4 predictors concordantly damaging |

### Evidence Summary

| Pathogenic | Benign |
|------------|--------|
| 0 Very Strong | None |
| 0 Strong | |
| 3 Moderate (PM2, PM5, PM1) | |
| 1 Supporting (PP3) | |

### Classification: **LIKELY PATHOGENIC**

**Rationale**: 3 Moderate + 1 Supporting meets ACMG criteria for Likely Pathogenic. Strong structural evidence at known hotspot position supports classification upgrade from VUS.

**Confidence**: ★★☆ (Moderate) - Structural evidence strong, but lacks direct R175C functional data

---

## 8. Clinical Recommendations

### For Affected Individual

| Category | Recommendation |
|----------|----------------|
| **Confirmation** | Consider orthogonal confirmation |
| **Clinical management** | Li-Fraumeni screening protocol |
| **Cancer surveillance** | Comprehensive multi-organ screening |
| **Genetic counseling** | Discuss uncertainty, manage expectations |

### Family Testing Considerations

- Likely Pathogenic classification supports cascade testing
- Counsel families about classification confidence level
- Consider research enrollment for functional studies

### Recommended Surveillance (Li-Fraumeni Protocol)

| Modality | Frequency | Starting Age |
|----------|-----------|--------------|
| Breast MRI | Annual | 20-25 |
| Colonoscopy | Every 2-5 years | 25 |
| Skin exam | Annual | 18 |
| Whole-body MRI | Consider annual | 25 |

---

## 9. Limitations & Uncertainties

| Factor | Impact |
|--------|--------|
| **No direct functional data** | Classification relies on position analogy |
| **ClinVar discordance** | Our Likely Pathogenic vs ClinVar VUS |
| **Phenocopy risk** | Environmental factors could explain cancer |

**Recommendation**: Consider this classification provisional pending functional studies. Reanalyze when new data available.

---

## Data Sources

| Tool | Purpose | Result |
|------|---------|--------|
| `ClinVar_search_variants` | Classification | VCV000376642 (VUS) |
| `gnomad_search_variants` | Population freq | 0 (absent) |
| `MyVariant_query_variants` | Predictions | All damaging |
| `RCSBData_get_entry` | Structure | 2OCJ |
| `PubMed_search_articles` | Literature | R175 studies |

---

Example 3: Likely Benign Common Variant

User Query

"What is the significance of the CFTR p.M470V variant?"

Expected Output (Key Sections)

# Variant Interpretation Report: CFTR p.M470V

## Executive Summary

| Field | Value |
|-------|-------|
| **Variant** | NM_000492.4:c.1408A>G (p.Met470Val) |
| **Gene** | CFTR |
| **Classification** | **BENIGN** ★★★ |
| **Evidence Strength** | Very Strong (population frequency) |
| **Key Finding** | Common variant (14% global AF); well-established benign polymorphism |

---

## 2. Population Data

### gnomAD v4.0 Frequencies

| Population | Allele Frequency |
|------------|------------------|
| **Overall** | 0.14 (14%) |
| European | 0.22 |
| East Asian | 0.03 |
| African | 0.08 |

**Interpretation**: Far too common to cause rare disease like CF (prevalence 1/3,000). **BA1 applies**.

---

## 7. ACMG Classification

### Evidence Codes Applied

| Code | Strength | Rationale |
|------|----------|-----------|
| **BA1** | Stand-Alone | AF >5% in gnomAD |
| **BS1** | Strong | AF far exceeds disease frequency |
| **BP4** | Supporting | Benign computational predictions |

### Classification: **BENIGN**

**Rationale**: BA1 alone is sufficient for Benign classification. This is a well-established benign polymorphism.

---

## 8. Clinical Recommendations

### For Individual
- This variant is **NOT** disease-causing
- Do not use for CF carrier screening interpretation
- No clinical follow-up needed based on this variant

### Important Notes
- May be found in compound heterozygosity with pathogenic variants
- Presence does NOT modify carrier status determined by other variants

---

Example 4: Novel Variant with Limited Data

User Query

"Interpret SCN5A c.4813+3A>G - novel splice variant"

Expected Output (Key Sections)

# Variant Interpretation Report: SCN5A c.4813+3A>G

## Executive Summary

| Field | Value |
|-------|-------|
| **Variant** | NM_000335.5:c.4813+3A>G |
| **Gene** | SCN5A |
| **Classification** | **VUS - FAVOR PATHOGENIC** ★☆☆ |
| **Evidence Strength** | Limited (splice prediction only) |
| **Key Finding** | Near-splice variant in cardiac arrhythmia gene; SpliceAI predicts donor loss |

**Action**: Functional RNA studies recommended before clinical use.

---

## 4. Computational Predictions

### Splice Predictions

| Tool | Score | Prediction |
|------|-------|------------|
| SpliceAI donor loss | 0.89 | High impact predicted |
| MaxEntScan (WT) | 8.2 | Strong donor site |
| MaxEntScan (Mut) | 4.1 | Weakened |

**Interpretation**: +3 position variants can affect splicing. SpliceAI score of 0.89 suggests high likelihood of splice disruption.

---

## 7. ACMG Classification

### Evidence Codes Applied

| Code | Strength | Rationale |
|------|----------|-----------|
| **PM2** | Supporting | Absent from gnomAD |
| **PP3** | Supporting | SpliceAI predicts splice disruption |
| **PM4** | Supporting | If causes in-frame deletion, alters protein |

### Classification: **VUS (Favor Pathogenic)**

**Rationale**: Insufficient evidence for Likely Pathogenic (would need 1 Strong + 1 Moderate or 3 Moderate). Splice prediction is suggestive but not confirmed.

---

## 8. Clinical Recommendations

### Recommended Studies
1. **RNA studies** - RT-PCR from patient sample to confirm aberrant splicing
2. **Family segregation** - Test affected relatives
3. **Functional studies** - Minigene assay if RNA unavailable

### Clinical Management
- Classification is VUS: Do NOT use for clinical decisions
- Continue phenotype-based cardiac management
- Offer participation in research studies

### Reclassification Potential
- RNA showing exon skipping → Upgrade to Likely Pathogenic (PVS1)
- Segregation in 2+ affected → Add PP1

Clinical Variant Interpreter - Tool Reference

Core Annotation Tools

MyVariant.info - Aggregated Annotations

Example - Query variant:

result = tu.tools.MyVariant_query_variants(
    variant_id="chr17:g.7674220C>T",
    fields="clinvar,gnomad,cadd,dbnsfp"
)
# Returns: ClinVar, gnomAD, CADD, dbNSFP predictions

Key Fields:

---

ClinVar - Clinical Classifications

Example - Search ClinVar:

result = tu.tools.ClinVar_search_variants(
    variant="NM_007294.4:c.5266dupC"
)
# Returns: VCV ID, classification, review status, submitters

Classification Interpretation:

---

VariantValidator - MANE Transcript Lookup & Variant Validation

Example - Get MANE transcript:

result = tu.tools.VariantValidator_gene2transcripts(
    gene_symbol="TP53", transcript_set="mane", genome_build="GRCh38"
)
# Returns: [{current_symbol: "TP53", transcripts: [{reference: "NM_000546.6",
#   annotations: {mane_select: true, mane_plus_clinical: false}}]}]

Example - Validate variant:

result = tu.tools.VariantValidator_validate_variant(
    genome_build="GRCh38",
    variant_description="NM_007294.4:c.5266dup",
    select_transcripts="NM_007294.4"
)
# Returns: validated HGVS, protein consequence, genomic coordinates, gene IDs

When to use:

• Phase 1: Always use gene2transcripts to identify the MANE Select transcript before annotating variants
• Phase 1: Use validate_variant to normalize user-provided HGVS notation and get cross-genome-build coordinates
• Prefer MANE Select transcript for canonical annotation; fall back to MANE Plus Clinical if relevant

---

gnomAD - Population Frequencies

Example - Query gnomAD:

result = tu.tools.gnomad_search_variants(
    variant="17-7674220-C-T"
)
# Returns: AF, ancestry-specific AFs, AC, AN, homozygotes

ACMG Frequency Thresholds:

Ancestry-Specific Populations:

---

ClinGen - Gene Validity & Dosage Sensitivity (NEW)

Authoritative curation of gene-disease relationships from ClinGen.

Gene Validity

Example - Check gene-disease validity:

result = tu.tools.ClinGen_search_gene_validity(gene="BRCA1")
# Returns: Classification (Definitive/Strong/Moderate/Limited), disease, inheritance

Validity Classification Interpretation:

Dosage Sensitivity

Example - Check haploinsufficiency:

result = tu.tools.ClinGen_search_dosage_sensitivity(gene="MECP2")
# Returns: Haploinsufficiency Score (0-3), Triplosensitivity Score (0-3)

Dosage Score Interpretation (for CNVs):

Clinical Actionability

Why ClinGen is Critical:

• Required for PP4 (phenotype specificity)
• Establishes gene-disease validity before classification
• Dosage scores critical for PVS1 in CNV interpretation
• Actionability informs return of incidental findings

---

SpliceAI - Splice Variant Prediction (NEW)

Deep learning model for predicting splice-altering effects.

Example - Predict splice effect:

# Full prediction
result = tu.tools.SpliceAI_predict_splice(
    variant="chr17-41276045-A-G",
    genome="38"
)
# Returns: DS_AG, DS_AL, DS_DG, DS_DL scores, max_delta_score, interpretation

# Quick triage
quick = tu.tools.SpliceAI_get_max_delta(
    variant="chr17-41276045-A-G"
)
# Returns: max_delta_score, interpretation

Variant Format: chr{chrom}-{pos}-{ref}-{alt} or {chrom}:{pos}:{ref}:{alt}

Delta Score Types:

Score Interpretation for ACMG:

When to Use:

• Intronic variants within ±50bp of splice sites
• Synonymous/missense variants (may still affect splicing)
• Deep intronic variants creating cryptic splice sites
• Validation when functional studies suggest splice defect

---

Pathogenicity Prediction Tools (NEW)

CADD - Combined Annotation Dependent Depletion (NEW API)

Example - Score a variant:

result = tu.tools.CADD_get_variant_score(
    chrom="17",
    pos=7674220,
    ref="G",
    alt="A",
    version="GRCh38-v1.7"  # Options: GRCh38-v1.7, GRCh37-v1.7
)
# Returns: phred_score, raw_score, interpretation

CADD PHRED Score Interpretation:

---

AlphaMissense - DeepMind Pathogenicity Prediction (NEW)

State-of-the-art deep learning model for missense pathogenicity prediction.

Example - Get pathogenicity score:

result = tu.tools.AlphaMissense_get_variant_score(
    uniprot_id="P00533",  # EGFR
    variant="L858R"  # or "p.L858R"
)
# Returns: pathogenicity_score, classification

AlphaMissense Thresholds (from Cheng et al., Science 2023):

Why Use AlphaMissense:

• ~90% accuracy on ClinVar pathogenic variants
• Covers all ~71 million possible human missense variants
• Trained on evolutionary data, not clinical annotations

---

ESMC-6B SAE - Mechanism of Effect (for VUS missense)

AlphaMissense scores tell you a variant is likely pathogenic. ESMC-6B Sparse Autoencoder features tell you which interpretable protein-language-model features the mutation disrupts — catalytic, ligand-binding, PTM, domain, transmembrane, etc. Use as a mechanism complement to pathogenicity scores.

Example — One-call mechanism for a VUS:

result = tu.tools.ESM_explain_variant_mechanism(
    sequence=wt_aa_sequence,
    position=600, ref_aa="V", alt_aa="E",
    top_k_features=5,
)
# result["data"]["mechanism_summary"] e.g.:
#   "Disrupted feature categories (lost): catalytic=2, ligand-binding=1"

Example — Saturation at one position (all 19 alternates):

from itertools import product
alts = "ACDEFGHIKLMNPQRSTVWY".replace(wt_residue, "")
variants = [{"position": 600, "ref_aa": "V", "alt_aa": a} for a in alts]
result = tu.tools.ESM_score_variant_sae_batch(
    sequence=wt_aa_sequence, variants=variants, top_k_features=5,
)
# Forge cost: 20 calls (1 ref + 19 mut), not 38 (2 per variant)

Mapping SAE categories → ACMG support:

Requires: ESM_API_KEY (free non-commercial token at https://forge.evolutionaryscale.ai) and pip install 'esm @ git+https://github.com/evolutionaryscale/esm@ee891c52' (PyPI esm 3.2.x lacks SAEConfig). Outputs governed by EvolutionaryScale Cambrian Inference License — non-commercial use only.

---

EVE - Evolutionary Variant Effect (NEW)

Unsupervised deep learning model using evolutionary data (Harvard/Oxford).

Example - Score variant:

# Via genomic coordinates
result = tu.tools.EVE_get_variant_score(
    chrom="17",
    pos=7674220,
    ref="C",
    alt="T"
)

# Or via HGVS
result = tu.tools.EVE_get_variant_score(
    variant="ENST00000269305.4:c.100G>A"
)
# Returns: eve_score, classification, gene, polyphen/sift from VEP

EVE Score Interpretation:

Note: EVE covers ~3,000 disease-related genes. Use EVE_get_gene_info to check coverage.

---

Integrating Prediction Tools

Best Practice for VUS Classification:

def get_multi_predictor_evidence(tu, variant_info):
    """
    Combine multiple predictors for robust PP3/BP4 assignment.
    """
    evidence = []
    
    # 1. CADD (all variants)
    cadd = tu.tools.CADD_get_variant_score(
        chrom=variant_info['chrom'],
        pos=variant_info['pos'],
        ref=variant_info['ref'],
        alt=variant_info['alt']
    )
    if cadd.get('status') == 'success':
        score = cadd['data']['phred_score']
        evidence.append({
            'tool': 'CADD',
            'score': score,
            'damaging': score >= 20
        })
    
    # 2. AlphaMissense (missense only)
    if variant_info.get('uniprot_id') and variant_info.get('aa_change'):
        am = tu.tools.AlphaMissense_get_variant_score(
            uniprot_id=variant_info['uniprot_id'],
            variant=variant_info['aa_change']
        )
        if am.get('status') == 'success' and am.get('data'):
            evidence.append({
                'tool': 'AlphaMissense',
                'score': am['data'].get('pathogenicity_score'),
                'classification': am['data'].get('classification'),
                'damaging': am['data'].get('classification') == 'pathogenic'
            })
    
    # 3. EVE (via VEP)
    eve = tu.tools.EVE_get_variant_score(
        chrom=variant_info['chrom'],
        pos=variant_info['pos'],
        ref=variant_info['ref'],
        alt=variant_info['alt']
    )
    if eve.get('status') == 'success':
        eve_scores = eve['data'].get('eve_scores', [])
        if eve_scores:
            evidence.append({
                'tool': 'EVE',
                'score': eve_scores[0].get('eve_score'),
                'damaging': eve_scores[0].get('eve_score', 0) > 0.5
            })
    
    # Consensus
    damaging = sum(1 for e in evidence if e.get('damaging'))
    benign = sum(1 for e in evidence if not e.get('damaging'))
    
    return {
        'predictions': evidence,
        'damaging_count': damaging,
        'benign_count': benign,
        'acmg_pp3': damaging >= 2 and benign == 0,
        'acmg_bp4': benign >= 2 and damaging == 0
    }

---

Somatic & Disease Association Tools (NEW)

COSMIC - Somatic Cancer Mutations

Example - Check if variant is somatic hotspot:

# Search for specific mutation
result = tu.tools.COSMIC_search_mutations(
    operation="search",
    terms="BRAF V600E",
    max_results=20
)
# Returns: mutation_id, cancer types, frequency

# Get all mutations for gene (hotspot analysis)
gene_muts = tu.tools.COSMIC_get_mutations_by_gene(
    operation="get_by_gene",
    gene="BRAF",
    max_results=200
)
# Returns: All mutations with cancer type distribution

COSMIC Evidence for ACMG:

OMIM - Mendelian Disease Context

⚠️ Requires: OMIM_API_KEY environment variable

Example - Get gene-disease context:

# Search for gene in OMIM
search = tu.tools.OMIM_search(
    operation="search",
    query="BRCA1",
    limit=5
)

# Get detailed entry with clinical info
entry = tu.tools.OMIM_get_entry(
    operation="get_entry",
    mim_number="113705"  # BRCA1
)

# Get clinical synopsis for phenotype matching
synopsis = tu.tools.OMIM_get_clinical_synopsis(
    operation="get_clinical_synopsis",
    mim_number="114480"  # Breast-ovarian cancer
)

OMIM Entry Types:

DisGeNET - Gene-Disease Associations

⚠️ Requires: DISGENET_API_KEY environment variable

Example - Get gene-disease evidence:

# Gene-disease associations
gda = tu.tools.DisGeNET_search_gene(
    operation="search_gene",
    gene="BRCA1",
    limit=20
)
# Returns: Associated diseases with scores

# High-confidence curated associations
curated = tu.tools.DisGeNET_get_gda(
    operation="get_gda",
    gene="BRCA1",
    source="CURATED",
    min_score=0.5
)

# Variant-disease associations
vda = tu.tools.DisGeNET_get_vda(
    operation="get_vda",
    gene="BRCA1",
    limit=30
)

DisGeNET Score for ACMG:

---

Regulatory Context Tools (NEW)

ChIPAtlas - Transcription Factor Binding

Example - Check TF binding at variant:

# Get TF binding near gene
tf_binding = tu.tools.ChIPAtlas_enrichment_analysis(
    gene="BRCA1",
    cell_type="all"
)
# Returns: TFs with binding peaks near gene

# Get specific peaks
peaks = tu.tools.ChIPAtlas_get_peak_data(
    gene="BRCA1",
    experiment_type="TF"
)

Use for: Non-coding variants that may disrupt TF binding sites

ENCODE - Regulatory Elements

Example - Get regulatory annotations:

# Search for regulatory data near variant
experiments = tu.tools.ENCODE_search_experiments(
    assay_title="ATAC-seq",
    biosample="heart"
)
# Returns: Open chromatin experiments

Key ENCODE Assays:

---

Expression Context Tools (NEW)

CELLxGENE - Single-Cell Expression

Example - Validate tissue expression:

# Get expression in disease-relevant tissue
expression = tu.tools.CELLxGENE_get_expression_data(
    gene="FBN1",
    tissue="heart"
)
# Returns: Expression per cell type (cardiomyocytes, fibroblasts, etc.)

Why use it: Confirms gene is expressed in phenotype-relevant cells

---

Literature Tools (ENHANCED)

BioRxiv/MedRxiv - Preprints

Example - Search preprints (bioRxiv/medRxiv don't have search APIs, use EuropePMC):

# Search for recent findings
preprints = tu.tools.EuropePMC_search_articles(
    query="BRCA1 variant functional",
    source="PPR",  # PPR = Preprints only
    pageSize=10
)

⚠️ Important: Always flag preprints as NOT peer-reviewed

OpenAlex - Citation Analysis

Example - Get citation counts:

# Get citation metrics for key paper
work = tu.tools.openalex_search_works(
    query="BRCA1 functional study pathogenic",
    limit=5
)
# Returns: Papers with cited_by_count, is_oa, etc.

Semantic Scholar - AI-Ranked Search

Example:

papers = tu.tools.SemanticScholar_search_papers(
    query="BRCA1 c.5266dupC pathogenic",
    limit=15
)

---

Ensembl - Variant Effect Predictor

Example - Get VEP data:

result = tu.tools.EnsemblVar_get_variant_consequences(
    variant_id="rs28934576"
)
# Returns: Consequence, transcript, SIFT, PolyPhen

---

Disease Association Tools

OMIM - Gene-Disease Relationships

Example - Get OMIM associations:

result = tu.tools.OMIM_search(query="BRCA1")
# Returns: MIM#, gene-phenotype relationships, inheritance

---

ClinGen - Gene Validity

Gene Validity Levels:

---

Structural Analysis Tools

PDB - Experimental Structures

Example - Get structure:

# Find PDB structures for TP53
hits = tu.tools.PDBe_get_uniprot_mappings(uniprot_id="P04637")
if hits:
    structure = tu.tools.PDB_get_structure(pdb_id=hits[0]['pdb_id'])

AlphaFold - Predicted Structures

Example - Get AlphaFold structure:

# From AlphaFold DB
structure = tu.tools.alphafold_get_prediction(accession="P04637")

# Or predict de novo
structure = tu.tools.NvidiaNIM_alphafold2(
    sequence=protein_sequence,
    algorithm="mmseqs2"
)

pLDDT Interpretation:

---

Domain/Function Tools

Example - Get domains:

domains = tu.tools.InterPro_get_protein_domains(accession="P04637")
# Returns: Domain boundaries, types, functions

---

Literature Tools

PubMed - Literature Search

Example - Search for functional studies:

# Gene + variant search
result = tu.tools.PubMed_search_articles(
    query="BRCA1 AND c.5266dupC",
    max_results=10
)

# Functional studies
result = tu.tools.PubMed_search_articles(
    query="BRCA1 AND functional study",
    max_results=20
)

Search Strategies:

---

Workflow Code Examples

Example 1: Complete Variant Annotation

def annotate_variant(tu, variant_hgvs, gene):
    """Complete variant annotation workflow."""
    
    # Phase 1: Get aggregated annotations
    annotations = tu.tools.MyVariant_query_variants(
        variant_id=variant_hgvs,
        fields="clinvar,gnomad,cadd,dbnsfp"
    )
    
    # Phase 2: ClinVar detail
    clinvar = tu.tools.ClinVar_search_variants(variant=variant_hgvs)
    
    # Phase 3: Population frequency
    gnomad = tu.tools.gnomad_search_variants(variant=variant_hgvs)
    
    # Phase 4: Gene context
    omim = tu.tools.OMIM_search(query=gene)
    
    # Phase 5: Literature
    literature = tu.tools.PubMed_search_articles(
        query=f"{gene} AND {variant_hgvs}",
        max_results=20
    )
    
    return {
        'annotations': annotations,
        'clinvar': clinvar,
        'gnomad': gnomad,
        'omim': omim,
        'literature': literature
    }

Example 2: Structural Analysis for VUS

def structural_analysis_for_vus(tu, gene, uniprot_id, residue_position):
    """Structural analysis for VUS missense variants."""
    
    # Try PDB first
    pdb_structures = tu.tools.PDBe_get_uniprot_mappings(uniprot_id=uniprot_id)
    
    if pdb_structures:
        # Use best resolution experimental structure
        best_pdb = sorted(pdb_structures, key=lambda x: x.get('resolution', 10))[0]
        structure = tu.tools.PDB_get_structure(pdb_id=best_pdb['pdb_id'])
        structure_source = f"PDB {best_pdb['pdb_id']}"
    else:
        # Fallback to AlphaFold
        structure = tu.tools.alphafold_get_prediction(accession=uniprot_id)
        structure_source = "AlphaFold DB"
    
    # Get domain information
    domains = tu.tools.InterPro_get_protein_domains(accession=uniprot_id)
    
    # Get functional sites
    functions = tu.tools.UniProt_get_function_by_accession(accession=uniprot_id)
    
    # Analyze residue context
    analysis = {
        'structure_source': structure_source,
        'domains': identify_domain(domains, residue_position),
        'functional_sites': find_nearby_sites(functions, residue_position),
        'pm1_applicable': assess_pm1(domains, functions, residue_position)
    }
    
    return analysis

Example 3: ACMG Classification

def calculate_acmg_classification(evidence_codes):
    """Calculate ACMG classification from evidence codes."""
    
    # Count evidence
    pathogenic = {
        'very_strong': [],
        'strong': [],
        'moderate': [],
        'supporting': []
    }
    benign = {
        'stand_alone': [],
        'strong': [],
        'supporting': []
    }
    
    for code, strength in evidence_codes:
        if code.startswith(('PVS', 'PS', 'PM', 'PP')):
            # Pathogenic evidence
            if strength == 'very_strong':
                pathogenic['very_strong'].append(code)
            elif strength == 'strong':
                pathogenic['strong'].append(code)
            elif strength == 'moderate':
                pathogenic['moderate'].append(code)
            else:
                pathogenic['supporting'].append(code)
        else:
            # Benign evidence
            if code == 'BA1':
                benign['stand_alone'].append(code)
            elif strength == 'strong':
                benign['strong'].append(code)
            else:
                benign['supporting'].append(code)
    
    # Apply ACMG rules
    if benign['stand_alone']:
        return 'Benign'
    
    if len(benign['strong']) >= 2:
        return 'Benign'
    
    vs = len(pathogenic['very_strong'])
    s = len(pathogenic['strong'])
    m = len(pathogenic['moderate'])
    p = len(pathogenic['supporting'])
    
    if (vs >= 1 and (s >= 1 or m >= 1 or m >= 2 or p >= 2)) or \
       (s >= 2) or \
       (s >= 1 and m >= 3):
        return 'Pathogenic'
    
    if (vs >= 1 and m >= 1) or \
       (s >= 1 and m >= 1 or m >= 2) or \
       (s >= 1 and p >= 2):
        return 'Likely Pathogenic'
    
    if len(benign['strong']) >= 1 and len(benign['supporting']) >= 1:
        return 'Likely Benign'
    
    return 'VUS'

---

Fallback Chains

Variant Annotations

Structure

Gene Information

Literature

---

Common Parameter Mistakes

---

ACMG Code Quick Reference

Pathogenic Codes

Benign Codes

---

Quality Thresholds

Computational Predictions (Updated)

Recommended Order of Use: 1. AlphaMissense (highest accuracy for missense, ~90%) 2. CADD (works for all variant types) 3. EVE (unsupervised, complements AlphaMissense) 4. SIFT/PolyPhen (legacy, for comparison)

Concordance for PP3/BP4

---

Rate Limits

Handle with appropriate delays between calls.