AlphaGenome ASO Workflow

This repository provides a reproducible workflow for antisense oligonucleotide (ASO) masking experiments using AlphaGenome. The analysis proceeds through:

1. Visualize the gene and context-specific model outputs
2. Generate ASO-masked sequence variants across a target window
3. Predict effects and compute ASO impact scores
4. Visualize ASO scores as sequence logos around the target exon

The alphagenome library is used as-is via its public API.

---

1. Installation

Get an AlphaGenome API key: https://deepmind.google.com/science/alphagenome/

Create and activate a Python environment (Python ≥3.10 recommended):

conda create -y -n alphagenome_env python=3.11
conda activate alphagenome_env
pip install -e alphagenome biopython pandas numpy matplotlib

---

2. Reference Data (GRCh38 / GENCODE v46)

Download primary assembly FASTA:

cd data
wget https://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_46/GRCh38.primary_assembly.genome.fa.gz
gunzip GRCh38.primary_assembly.genome.fa.gz
cd ..

Note: AlphaGenome currently supports GENCODE up to v46.

---

3. Configuration

Edit parameters (e.g., config_private/SMN2.json) before running aso.ipynb:

• gene_symbol: Target gene (e.g., SMN2)
• dna_api_key: AlphaGenome API key
• ontology_terms: Context identifiers (UBERON/CL ontology). See results/metadata*.csv
• requested_outputs: Prediction types (e.g., RNA_SEQ, SPLICE_SITE_USAGE)
• exon_intervals: Genomic start/end of target exon
• flank: Bases around exon to include in ASO window (<250 recommended)
• ASO_length: Length of the ASO masking window
• strand: Track strand filter (+, -, stranded, unstranded, all)
• track_filter: Substring to filter tracks (optional)
• SNV: Single nucleotide variant positions (0-based index; currently SNV is the only supported variant type).

Tips:

• Ensure results_dir exists or let the notebook create it; ASO outputs are saved under results_dir/ASO/.
• For SNVs, set position relative to the resized interval start (0-based); the notebook converts to 1-based for AlphaGenome.

---

4. How it works — ASO

ASO experiments mask a short window (length = ASO_length) with N bases while sliding across a target region around the exon. For each masked variant, AlphaGenome predicts selected outputs (e.g., RNA-seq coverage, splice-site usage). Differences relative to the reference are aggregated into ASO impact scores and visualized as sequence logos over the window.

Outputs

• ${results_dir}/ASO/{config}_ASO_scores.csv: Per-ASO scores (one row per mask position) for each requested output type.
• ${results_dir}/ASO/{config}_ASO_{OUTPUT}.bed: Top and bottom ASOs colored by effect size for genome browser tracks.
• ${results_dir}/ASO/{config}_ASO_{OUTPUT}_full.bed: Full set of ASO windows (all positions) for the output type.
• SeqLogo plots and overlaid tracks are rendered inline in aso.ipynb for quick inspection.