Applications — From Discovery to Therapeutics¶

This directory (docs/applications/) contains domain-specific workflows and use cases for Agentic-SpliceAI. Each application demonstrates how the multi-layer pipeline translates computational predictions into biological and clinical insights. As applications mature, they graduate to docs/products/ for production deployment guides.

The Translational Pathway¶

Current State: Limited by Canonical Annotations¶

Drug discovery today is constrained by incomplete gene annotations:

Known Genes (20,000)
    |
Canonical Isoforms (~20,000 in MANE)
    |
Druggable Targets (~3,000)
    |
Approved Drugs (~1,500 targets)

Bottleneck: 90% of splice sites are non-canonical. Current annotations capture only the most common isoforms, leaving the vast majority of biologically active splice variants undiscovered.

Expanding the Druggable Genome¶

Agentic-SpliceAI discovers context-specific isoforms to expand the therapeutic space. Consider BRCA1 as a concrete example:

Scope	Splice Sites	Isoforms	Targeting Strategy
Canonical (MANE)	44	1-2 main	Target "BRCA1" broadly
Full potential (Ensembl + Discovery)	1,218+ (27x more)	100+ (tissue/disease-specific)	Target tumor-specific isoforms selectively

By moving beyond canonical annotations, we unlock 10-100x more therapeutic targets with the potential for reduced toxicity through isoform-selective drug design.

Application Domains¶

Note: CLI commands shown in the workflows below reflect planned Phase 7-9 functionality. Current capabilities cover base layer prediction, feature engineering, and meta layer training (Phases 1-6).

1. Oncology: Cancer-Specific Isoform Targets¶

Challenge: Tumors express aberrant splice isoforms that drive proliferation, evade apoptosis, and resist therapy. These cancer-specific isoforms are invisible to standard genomic pipelines.

Workflow:

# Predict splice sites for BRCA1 with cancer context
agentic-spliceai-predict --genes BRCA1 --base-model openspliceai

# Discover cancer-specific isoforms (Phase 9)
agentic-spliceai discover --genes BRCA1 \
  --context cancer:breast \
  --evidence-sources gtex,tcga,clinvar

# Validate with agentic layer (Phase 7)
agentic-spliceai validate --isoforms output/brca1_novel.parquet \
  --agents literature,expression,clinical

Impact: Identify tumor-specific BRCA1 isoforms for isoform-selective inhibitors that spare normal tissue function.

2. Clinical Genetics: VUS Interpretation¶

Challenge: Thousands of variants of uncertain significance (VUS) in clinical sequencing may affect splicing but lack functional evidence for classification.

Workflow:

# Analyze splice impact of TP53 variants (Phase 8)
agentic-spliceai variant --vcf patient_variants.vcf \
  --genes TP53 \
  --clinvar-cross-reference

# Generate clinical report (Phase 7)
agentic-spliceai report --variants output/tp53_splice_impact.parquet \
  --format clinical \
  --evidence-level acmg

Impact: Reclassify VUS variants by quantifying their splice-disrupting potential with multi-source evidence, enabling actionable clinical decisions.

3. Neurology: Brain-Specific Isoforms¶

Challenge: The brain has the highest rate of alternative splicing of any tissue, with many neurological disorders linked to isoform dysregulation. Autism risk genes alone have hundreds of brain-specific splice variants.

Workflow:

# Discover brain-specific isoforms for autism risk genes (Phase 9)
agentic-spliceai discover \
  --genes SHANK3 NRXN1 SYNGAP1 CHD8 \
  --context tissue:brain \
  --evidence-sources gtex,brainspan

# Cross-reference with known autism associations (Phase 7)
agentic-spliceai validate --isoforms output/asd_novel.parquet \
  --agents literature,expression,conservation

Impact: Map the brain-specific isoform landscape of autism risk genes to identify therapeutic targets for splice-modulating interventions.

4. Drug Development: Isoform-Selective Therapeutics¶

Challenge: Most drugs target proteins without distinguishing between isoforms, leading to off-target effects when disease-relevant and normal isoforms are both inhibited.

Workflow:

# Identify druggable isoform-specific regions (Phase 9)
agentic-spliceai druggability --genes BCL2 MCL1 \
  --isoform-catalog output/novel_isoforms.parquet \
  --assess-selectivity

# Generate drug target report (Phase 7)
agentic-spliceai report --targets output/druggable_isoforms.parquet \
  --format pharma \
  --include-structure-prediction

Impact: Enable isoform-selective drug design that targets disease-specific variants while sparing normal protein function, reducing toxicity and improving therapeutic windows.

5. Biomarker Discovery: Liquid Biopsy¶

Challenge: Current liquid biopsy approaches rely on known mutations and methylation patterns. Disease-specific splice isoforms detectable in circulating RNA represent an untapped biomarker source.

Workflow:

# Discover tissue-specific isoform biomarkers (Phase 9)
agentic-spliceai discover --gene-list data/cancer_panel.txt \
  --context cancer:lung \
  --biomarker-mode \
  --min-tissue-specificity 0.8

# Validate biomarker candidates (Phase 7)
agentic-spliceai validate --isoforms output/biomarker_candidates.parquet \
  --agents expression,clinical \
  --cross-tissue-comparison

Impact: Identify cancer-specific splice isoforms as novel liquid biopsy biomarkers for early detection, treatment monitoring, and companion diagnostics.

Last Updated: March 2026