v0.5.0

First release since v0.3.2. Adds reproducible epitope database construction and a
benchmarking utility, and renames the pipeline executable.

Breaking changes

The command is now antigensoup, not daedalus. The conda environment,
install paths and user-facing messages were renamed to match. Existing installs
need reinstalling; see the updated instructions in the README.

Was	Now
daedalus	antigensoup
conda_ymls/daedalus_env.yml	conda_ymls/antigensoup_env.yml
conda env daedalus	conda env antigensoup
share/daedalus/	share/antigensoup/

This also fixes the install instructions, which already documented the new names
while the wrapper still looked for the old ones, so a clean install previously
failed.

The bundled iedb.fasta.gz has been removed. Build the database instead:

Rscript scripts/build_iedb_db.R --outdir databases/iedb

The bundled file carried no record of how it had been filtered, and was
demonstrably incomplete: it contained no 12-residue peptides at all, despite IEDB
holding 267,718 of them. That is the signature of two separate length queries,
covering 8-11 and 13-25, being concatenated. About 94% of its contents were MHC
ligands with no T-cell or B-cell evidence behind them.

It remains retrievable for reproducing older results:

git show 396d94b:iedb.fasta.gz > iedb.fasta.gz

New: IEDB database builder

scripts/build_iedb_db.R builds the epitope database directly from the current
IEDB export in a single command, and writes a FASTA, a metadata TSV and a build
report recording the release used and every filter applied.

Against the 2026-09-08 IEDB release: 2,752,497 epitope records in, 240,971 unique
peptides out. Defaults keep linear 8-25mer peptides over the 20 standard amino
acids with at least one positive experimental T-cell or B-cell assay, excluding
analogues, mimotopes, peptides only ever seen chemically modified, and
low-complexity sequences. Complexity thresholds were calibrated against the real
IEDB distribution rather than assumed.

Requires R with data.table, plus curl and unzip. A build takes about five
minutes and caches the IEDB archives for reuse.

Documented in docs/iedb_database_build.md.

New: spike-in benchmark generator

scripts/generate_spikein_benchmark.R builds controlled benchmarks for measuring
detection sensitivity and specificity. Artificial peptides are composition-matched
shuffles of real retained epitopes, proven absent from both the database and the
target proteins before being inserted at recorded positions, so a miss is a true
false negative rather than a guess.

Validated against the E. coli K-12 reference proteome: 1000 spikes over 4,403
proteins, 13 of 13 internal checks passed, and 1000 of 1000 recovered at the
correct protein and coordinate. The same epitopes give zero hits against the
unspiked proteome.

Requires R with Biostrings and data.table. Documented in
docs/spikein_benchmark.md, including the full
validation record and an explicit statement of what the benchmark does not test:
only the epitope-matching stage is exercised, not assembly or gene prediction.

Also included

From v0.3.3, which was never released: scripts/gen_variants.py for generating
Hamming-distance variants of specific epitopes, and --scaffolds input to skip
straight to gene prediction and epitope search.

CITATION.cff has been corrected; it still named the project Daedalus and pointed
at the old repository.

Note on assets

No database file is attached. The point of this release is that the database is
built from a named IEDB release with a recorded filter set, rather than shipped as
a snapshot whose provenance cannot be reconstructed.

Daedalus v0.3.2.

Daedalus is a metagenomic analysis pipeline that assembles shotgun sequencing data, predicts microbial genes, and identifies exact matches to known immune epitopes. It is designed to detect potential T cell cross-reactive peptides.

Highlights

• Added automatic skipping of completed pipeline steps (resume-safe execution)
• Bundled filtered human-relevant IEDB epitope database for convenience
• Improved installation layout (conda YAMLs, scripts directory)
• Expanded and clarified documentation

New features

• Daedalus now detects existing outputs and avoids re-running:
  • SRA download
  • Host read removal
  • Assembly
  • Gene prediction
• This makes Daedalus robust to job preemption and incremental reruns on HPC systems

Database

• Includes a filtered FASTA of human / human-associated epitopes from IEDB (late 2025)
• Provided for convenience only — users should cite IEDB and original sources

Notes

• Designed for paired-end Illumina shotgun metagenomic data
• Assumes human host for read removal