endophynd

Recover and classify fungal rDNA (ITS, LSU, SSU) from Logan/SRA data at the read level, producing QIIME2-compatible feature tables and metabarcoding-style reports.

Quick start

mamba env create -f envs/base.yml
conda activate endophynd
pip install -e .
endophynd --help
snakemake --configfile workflow/config/params.yml --cores 1 --dry-run

Design

See endophynd_development_plan.md for the full design and roadmap.
See docs/decisions.md for the append-only decision log.

Phase status

Phases were built out of order: the targeted-search engine (Phase 4) and the SRA streaming path landed early because a concrete application drove them. Calibration, multi-locus classification, and the reporting/benchmark phases remain open. Status is deliberately conservative — see docs/decisions.md for what each entry actually claims.

• Phase 0 — scaffold, config, cache manager, CLI skeleton, fixtures
• Phase 1 — read-level discovery MVP (Logan unitigs, streaming) — functional; validated on real Logan data (ERR15383529)
• Phase 1.5 — calibration map + mock community (not started — gating thresholds are still post-hoc/uncalibrated)
• Phase 2 — multi-locus classification + HTML report (not started)
• Phase 3 — local raw-read / SRA streaming recovery (Illumina + platform tiers) — source=sra path landed in the triage and retrieve_and_bait rules (D21/D24/D25); the live SRA stream still needs end-to-end validation
• Phase 3.5 — Logan vs raw SRA concordance benchmark (not started)
• Phase 4 — targeted mode (query-as-reference) — endophynd target engine built (D27): both aligners, BioProject expansion, 23 passing tests, re-found the RPB2 marker on real Logan data. First application is a 10-dataset GBI pilot scan (results/alternaria_vs_gbi10/, D28) — hypothesis-generating, not a validated assay; the SRA-source branch of target is not yet validated on a live stream
• Phase 5 — long-read paths (ONT/PacBio quality tiers defined in D25; paths not built)
• Phase 6 — Streamlit GUI
• Phase 7 — cloud scale-out (deferred)