terravic/alphagenome-variant-visualization-skill

AlphaGenome Variant Visualization & Interpretation Skill

This skill provides an automated workflow to predict, interpret, and visually present the functional regulatory consequences of human genetic variants across 11 modalities (RNA-seq, DNase-seq, ChIP-seq H3K27ac/H3K4me3/CTCF/TF, Splicing Site Usage, Splicing Junctions, CAGE, PRO-CAP).

It compiles findings into static analysis reports and generates self-contained, interactive HTML5 and WebGL visual dashboards (ui/alphagenome_dashboard.html).

---

1. Skill Execution Workflow

Follow this multi-step procedure when processing user variant queries:

Step 1: Gene and Transcript Resolution

Resolve target gene coordinates, canonical transcript identifier, strand, and exon structure:

python3 scripts/lookup_gene_info.py --gene <GENE_SYMBOL> --pretty
# Or query by genomic position:
python3 scripts/lookup_gene_info.py --coord <CHROM:POS> --pretty

Step 2: Biosample and Ontology Resolution

Map the target tissue, disease context, or cell type to standard ontology identifiers (UBERON, CL, EFO):

python3 scripts/resolve_ontology_terms.py --term "<TISSUE_OR_CELL_TYPE>" --pretty

Step 3: Splicing Interpretation

When evaluating variants in canonical splice sites (+1/+2 donor, -1/-2 acceptor), exonic splice enhancers/silencers, or deep intronic cryptic sites:

python3 scripts/interpret_splicing.py --benchmark <BENCHMARK_CASE> --ascii
# Or with custom JSON:
python3 scripts/interpret_splicing.py --input <DATA_JSON> --ascii

Step 4: In-Silico Mutagenesis & TF Motif Analysis

Assess sequence importance around the variant (+/-20bp), evaluate position weight matrix (PWM) motif disruptions (e.g. CTCF, SP1, GATA1, TAL1, FOXA1), and calculate substitution matrices:

python3 scripts/analyze_ism.py --benchmark <BENCHMARK_CASE> --ascii
# Or with explicit sequence:
python3 scripts/analyze_ism.py --seq "<40BP_SEQUENCE>" --pos 20 --ref <REF_BASE> --alt <ALT_BASE> --ascii

Step 5: Epigenomic Signal and Track Profiling

Generate signal curves for RNA-seq, DNase accessibility, H3K27ac, and H3K4me3:

python3 scripts/visualize_genome_tracks.py --benchmark <BENCHMARK_CASE> --output-plot tracks.png
python3 scripts/visualize_variant_effects.py --benchmark <BENCHMARK_CASE> --output-radar radar.png

Step 6: Interactive Dashboard Compilation

Compile the comprehensive analysis dataset into the standalone interactive HTML5 dashboard:

python3 scripts/generate_dashboard.py \
  --input tests/synthetic_benchmarks.json \
  --case <BENCHMARK_CASE> \
  --output ui/alphagenome_dashboard.html

---

2. Standard Synthetic Benchmark Cases

The skill includes four fully synthetic test cases with zero patient health information (PHI):

1. syn_COL6A2_skip:
   • Variant: chr21:46126238:G>C
   • Gene: COL6A2 (Exon 3 canonical 5' splice donor disruption)
   • Biosample: Skeletal Muscle (UBERON:0001134)
   • Consequence: Complete Exon 3 skipping ($\Delta\Psi = -0.960$), U1 splice donor consensus ablated, -65% transcript loss via nonsense-mediated decay (NMD).
2. syn_TP53_promoter:
   • Variant: chr17:7675148:G>A
   • Gene: TP53 (Core promoter -85bp from TSS)
   • Biosample: Liver / Hepatocyte (UBERON:0002107)
   • Consequence: Collapse of open chromatin (-88% DNase), CTCF/SP1 motif ablation, -85% RNA expression drop.
3. syn_HBA2_enhancer:
   • Variant: chr16:223500:T>G
   • Gene: HBA2 (Distal intergenic regulatory region +24kb)
   • Biosample: Erythroblasts / K562 (CL:0000765)
   • Consequence: De novo GATA1/TAL1 enhancer creation, +140% erythroid-specific globin expression boost.
4. syn_ACTB_control:
   • Variant: chr7:5529776:C>T
   • Gene: ACTB (Exon 4 synonymous variant)
   • Biosample: Ubiquitous / HepG2 (EFO:0001187)
   • Consequence: Neutral baseline control, zero splicing or epigenetic change ($\Delta\Psi = 0.00$).

---

3. Reference Documentation

For detailed technical standards and schema specifications, consult the reference documents:

• AlphaGenome API Reference: Modality definitions, coordinate indexing, and JSON schemas.
• Variant Interpretation Guide: Statistical cutoffs, decision rules, and NMD logic.
• Report Templates: Standard research and clinical reporting formats.
• Dashboard Guide: Specifications for the six interactive dashboard panels.

---

4. Example Prompts for Agent Harnesses

Sample Prompt 1 (Splicing Analysis)

"Analyze the variant chr21:46126238 G>C in COL6A2 in skeletal muscle. Quantify the splicing impact, check for exon skipping or nonsense-mediated decay, and generate the interactive visual dashboard."

Sample Prompt 2 (Promoter Disruption)

"Evaluate the regulatory consequence of chr17:7675148 G>A located in the TP53 promoter in liver tissue. Assess transcription factor motif loss, chromatin accessibility changes, and produce the multi-track visual report."

Sample Prompt 3 (De Novo Enhancer Gain)

"Screen the intergenic variant chr16:223500 T>G near HBA2 in erythroblasts. Identify if any de novo transcription factor motifs are created and generate the full 6-panel interactive UI dashboard."