Phase 1 Design

Assumptions

  • Public studies will vary in raw-data availability, so AREE supports raw-data reanalysis and processed-results harmonization.
  • The first production species is Crassostrea gigas, but evidence records include species and ortholog fields from the start.
  • Identifier mappings are versioned data products with confidence levels, not silent replacements.
  • Phenotype labels need controlled terms while preserving original study descriptions.
  • Candidate ranking must be transparent and tunable, not a black-box model.

Architecture

AREE has five connected layers:

  1. Study registry and dataset intake.
  2. Standardized raw-data reanalysis (Nextflow RNA-seq pipeline).
  3. Cross-study evidence harmonization.
  4. Meta-analysis and candidate prioritization.
  5. User-facing reports and interface.

Provenance Model

Each evidence row retains study ID, source accession, input file, checksum, workflow version, software method, genome and annotation version, generation date, quality flags, and curation caveats.

Ranking Framework

The MVP candidate score combines study count, biological sample size, effect magnitude, adjusted significance, direction consistency, phenotype relevance, tissue/life-stage breadth, assay diversity, mapping confidence, quality flags, and heterogeneity penalty.