Skip to content

Applications

Public ledger of matured application bundles for Agentic-SpliceAI.

An application is a refined, packaged bundle of use cases that represents a user-facing functionality aligned with project goals. Each entry below curates a subset of example scripts under examples/ into a coherent unit, declares the src/ surface it depends on, and records its current maturity.

For the biology and translational-impact narrative (oncology, clinical VUS, neurology, drug development, biomarkers), see use_cases.md.

Project goals

Agentic-SpliceAI's applications serve a hierarchy of goals:

  1. Adaptive splice prediction — going beyond canonical annotations with context-aware models
  2. Novel isoform discovery — identifying the 90% of splice sites outside MANE/RefSeq
  3. Drug target identification — translating isoform discoveries into therapeutically actionable targets

Each application below declares which goal(s) it serves.

Maturity tiers

Applications live in one of four tiers. For the full definitions, graduation signals, and demotion triggers, see the methodology doc at dev/system_design/maturity_lifecycle.md.

Tier Meaning
Incubating 1-2 scripts, proof of concept; no ledger entry typically
Active Multi-script workflow, results folder, benchmarks; ledger entry exists
Mature Stable driver, reproducible benchmarks, downstream consumers
Experimental Sub-project with its own lifecycle (e.g., foundation models)
Product Deployable commitment (see ../products/ — empty)

Ledger

# Application Goals served Tier Driving examples Spec
1 Canonical Splice Prediction Adaptive prediction Mature examples/base_layer/ canonical_splice_prediction/
2 Adaptive Splice Prediction (M1/M2) Adaptive prediction Active examples/meta_layer/ adaptive_splice_prediction/
3 Multimodal Feature Engineering Cross-cutting Active examples/features/ multimodal_features/
4 Genomic Data Preparation Cross-cutting Active examples/data_preparation/ genomic_data_preparation/
5 Variant Effect Analysis (M4) Adaptive + drug target Active examples/variant_analysis/ variant_analysis/
6 Bioinformatics Lab UI All goals Mature server/bio/, notebooks/bioinfo_ui/ bioinformatics_lab_ui/
7 Novel Isoform Discovery (M3) Isoform discovery Incubating (planned) novel_isoform_discovery/
8 Agentic Validation All goals Incubating examples/agentic_layer/ agentic_validation/
9 Foundation Model Predictors Adaptive prediction Experimental examples/foundation_models/ foundation_model_predictors/

Reading the ledger: applications are named by user-facing functionality, not by examples/<topic>/ folder name. One topic can seed multiple applications; one application can pull from multiple topics.

How to read a spec

Each application spec (docs/applications/<name>/README.md) follows the same structure, adapted from the template at _template.md:

  • Problem — what biological or computational question is being answered?
  • User-facing functionality — what does this application let a user do?
  • Goals served — which of the three project goals does this advance?
  • Driving examples — the curated list of examples/<topic>/*.py scripts that demonstrate the application
  • src/ surface — the library modules the application relies on
  • Evaluation — benchmarks, baselines, metrics; links to results folders
  • Maturity tier — current tier with the signals that support it
  • Graduation signals — what would move this to the next tier
  • Known limitations — scope boundaries, failure modes

Adding a new application

An application is created when an examples/<topic>/ reaches late-Active maturity (see dev/system_design/maturity_lifecycle.md): the topic has multiple scripts, reproducible benchmarks, a canonical driver, and a clear user-facing functionality.

Process:

  1. Draft an internal harvest note at dev/applications/<name>-harvest.md scoping the application
  2. Copy _template.md to docs/applications/<name>/README.md
  3. Fill in each section, linking to driving examples, src/ modules, and results
  4. Add a row to the ledger table above
  5. Cross-link from any related tutorials under docs/tutorials/

Promotion to product

No application is currently being promoted. Product promotion requires all six criteria under dev/system_design/maturity_lifecycle.md#tier-4--product to be demonstrably met, and a deliberate decision to take on maintenance.

The products catalog lives at ../products/ and is currently empty.

See also