giae 0.2.2

Genome Interpretation & Annotation Engine - An Explainable, Evidence-Centric Framework for Genomic Interpretation

GIAE — Genome Interpretation & Annotation Engine

Explainable, evidence-first genome annotation. Every prediction shows its reasoning.

---

GIAE is a genome annotation engine that shows its work. Where Bakta and Prokka return labels, GIAE returns labels with the full evidence stack, a numerical confidence score, the reasoning chain that produced it, and the alternatives it considered.

It's also faster and more accurate than Bakta on the genomes we've benchmarked.

┌─────────────────────────────────────────────────────────────────────────┐
│  Phage benchmark — same FASTA, same scoring, same overlap criterion    │
├──────────────┬──────────────┬──────────────┬──────────────┬─────────────┤
│   Genome     │   GIAE F1    │   Bakta F1   │     Δ        │   Verdict   │
├──────────────┼──────────────┼──────────────┼──────────────┼─────────────┤
│   phiX174    │   60.0 %     │   60.0 %     │     —        │   tied      │
│   λ phage    │   79.2 %     │   72.6 %     │   +6.6 %     │   GIAE wins │
│   T7         │   88.1 %     │   85.2 %     │   +2.9 %     │   GIAE wins │
└──────────────┴──────────────┴──────────────┴──────────────┴─────────────┘
GIAE config: pyrodigal + rescue + phage_mode (no UniProt/BLAST/InterPro).
Bakta config: light db, --skip-* for missing tools, CDS pipeline active.

Reproduce these numbers →

---

Why GIAE

	Prokka / Bakta / RAST	GIAE
Output per gene	Label only	Label + evidence chain + confidence score + alternatives
Uncertainty	Hidden	Explicit, numeric, calibrated
Conflicting evidence	Silently resolved	Surfaced and reported
"hypothetical protein"	End of the line	Ranked as research priority with suggested experiments
Reasoning audit	None	Full reasoning chain in every record
Accuracy on phages	Baseline	Matches or beats Bakta on all 3 reference genomes
Deployment	CLI only	CLI + Python library + REST API + Docker stack

---

Architecture at a glance

              ┌─────────────────────────────────────────────────────────┐
  Genome ───▶ │ pyrodigal ORF prediction                                │
  (.gb/.fa)   │  + ShortOrfRescue (RBS + codon-usage gate)              │
              │  + NestedOrfFinder (phage_mode, position-weighted SD)   │
              │  + Aragorn (tRNA) + Barrnap (rRNA)                      │
              └────────────────────────────┬────────────────────────────┘
                                           ▼
              ┌─────────────────────────────────────────────────────────┐
              │ Evidence extraction (typed, weighted, with provenance)  │
              │  • PROSITE motifs (1,298 patterns, bundled)             │
              │  • Pfam / HMMER domains (local pyhmmer)                 │
              │  • Diamond / BLAST+ homology (local DB)                 │
              │  • UniProt + InterPro (online, cached)                  │
              │  • ESM-2 protein language model (optional)              │
              │  • GenBank curator annotations                          │
              └────────────────────────────┬────────────────────────────┘
                                           ▼
              ┌─────────────────────────────────────────────────────────┐
              │ Aggregation → Hypothesis generation → Confidence score  │
              │ Tiered: GenBank ▶ Homology ▶ Domain ▶ Motif ▶ Combined  │
              │ Conflict detection, novelty scoring, dark-matter rank   │
              └────────────────────────────┬────────────────────────────┘
                                           ▼
              ┌─────────────────────────────────────────────────────────┐
              │ FunctionalAnnotator (COG category, GO terms, normalised │
              │ product) → Interpretation object with reasoning chain   │
              └────────────────────────────┬────────────────────────────┘
                                           ▼
                Markdown · JSON · Interactive HTML report · REST API

Full architecture doc →

---

Install

From PyPI

pip install giae

From source (recommended for dev)

git clone https://github.com/Ayo-Cyber/GIAE.git
cd GIAE
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

Optional capabilities

pip install "giae[annotation]"     # pyrodigal — required for ORF prediction on FASTA
pip install "giae[hmmer]"          # pyhmmer — local Pfam domain search
pip install "giae[ai]"             # ESM-2 + torch — protein LM embeddings
pip install "giae[api]"            # FastAPI + Celery + Postgres — REST API server

Requirements: Python 3.9+, BioPython, Click, Rich. PROSITE patterns are bundled — no database setup needed for the base pipeline.

---

Quickstart

CLI — interpret a genome

# Offline pipeline (PROSITE only, ~seconds)
giae interpret lambda_phage.gb

# Phage genome with nested-ORF detection
giae interpret phiX174.gb --phage

# Full online pipeline (UniProt + InterPro)
giae interpret lambda_phage.gb --mode online

# Interactive HTML report
giae interpret lambda_phage.gb --format html -o report.html

# Parallel workers for large genomes
giae interpret big_genome.gb --workers 8 --mode local

Python library

from giae.engine.interpreter import Interpreter
from giae.parsers.genbank import GenBankParser

genome = GenBankParser().parse("lambda_phage.gb")

interpreter = Interpreter(
    use_uniprot=False,         # offline mode
    use_interpro=False,
    phage_mode=True,           # phage-aware nested ORF detection
)
summary = interpreter.interpret_genome(genome)

for result in summary.results:
    interp = result.interpretation
    if not interp:
        continue
    print(f"{result.gene_id}: {interp.hypothesis}")
    print(f"  confidence: {interp.confidence_level.value} ({interp.confidence_score:.2f})")
    print(f"  COG:        {interp.metadata.get('cog_category')} — {interp.metadata.get('cog_name')}")
    print(f"  reasoning:  {' → '.join(interp.reasoning_chain)}")

REST API — full stack via Docker

cp .env.example .env  # then fill in JWT_SECRET, NEXTAUTH_SECRET
docker compose up -d postgres redis api worker
curl http://localhost:8000/api/v1/health

# Sign up, get a token, submit a genome
TOKEN=$(curl -sS -X POST http://localhost:8000/api/v1/auth/signup \
  -H "Content-Type: application/json" \
  -d '{"email":"[e-mail]","password":"correct-horse-battery"}' \
  | jq -r .access_token)

curl -X POST http://localhost:8000/api/v1/jobs \
  -H "Authorization: Bearer $TOKEN" \
  -F "file=@lambda_phage.gb" \
  -F "phage_mode=true"

Full REST API reference →

---

Example output

A well-characterised gene with converging evidence:

─────────────────────────────────────────────────────────────────
Gene: J  ·  Locus: lambda_J  ·  Length: 1,131 aa
Hypothesis:   Tail fiber / host-receptor binding protein
Confidence:   HIGH   (0.87)
COG:          X — Mobilome: prophages, transposons   (source: pfam)
Pfam:         PF09255 (Phage_tail_fib)
GO:           GO:0019028, GO:0019068

Evidence:
  [0.94] HMMER         Pfam_PF09255  e-value 2.1e-14
  [0.90] UniProt       P03722 — Tail fiber protein J, λ phage
  [0.82] PROSITE       PS51123 — Phage tail fiber repeat

Reasoning:
  1. Pfam domain hit Phage_tail_fib (e-value 2.1e-14) is diagnostic
  2. UniProt homolog P03722 is a Swiss-Prot reviewed entry from same organism
  3. PROSITE motif corroborates structural prediction

Uncertainty sources: none
Competing hypotheses: none above 0.50 threshold
─────────────────────────────────────────────────────────────────

A dark-matter gene flagged for research:

─────────────────────────────────────────────────────────────────
Gene: orf_b72b1  ·  147 aa
Interpretation:  NONE
Novelty class:   DARK MATTER  (no signal from any source)
Priority:        HIGH
Suggested experiments:
  • Recombinant expression and biochemical activity screening
  • Deletion mutant phenotyping for essentiality
  • Comparative genomics across related strains
  • Structural characterization (cryo-EM or AlphaFold + Foldseek)
─────────────────────────────────────────────────────────────────

---

Confidence model

Every prediction carries a numeric score [0.0, 1.0] mapped to a named level:

Level	Score	Meaning
HIGH	≥ 0.80	Multiple independent evidence types converge
MODERATE	0.50 – 0.79	Some convergence, or one strong source
LOW	0.30 – 0.49	Weak / single signal — treat as a lead
SPECULATIVE	< 0.30	Minimal signal, flagged for review
NONE	—	No evidence at all (dark matter)

Scoring penalties and bonuses are explicit and documented: architecture.md → Confidence model

---

What's in the box

Capability	Module	Status
ORF prediction (pyrodigal)	analysis/orf_finder.py	Always on
Short-ORF rescue (RBS + codon usage gate)	analysis/short_orf_rescue.py	Default on
Nested-ORF detection (phage mode)	analysis/nested_orf_finder.py	--phage
tRNA / tmRNA detection	analysis/aragorn.py	If aragorn on PATH
rRNA detection	analysis/barrnap.py	If barrnap on PATH
PROSITE motif scan	analysis/motif.py	Always on (1,298 patterns bundled)
Local Diamond BLASTP	analysis/diamond.py	If diamond on PATH + DB
Local NCBI BLAST+	analysis/blast_local.py	Fallback if Diamond absent
Local Pfam HMMER	analysis/hmmer.py	If pfam.hmm present
UniProt API client	analysis/uniprot.py	--mode online
InterPro / EBI HMMER client	analysis/interpro.py	--mode online
ESM-2 protein language model	analysis/ai.py	If [ai] extras installed
COG / GO functional annotation	analysis/functional_annotator.py	Always on (~100 Pfam IDs bundled)
Product-name normaliser	analysis/product_normalizer.py	Always on
HTML report generator	output/html_report.py	--format html
REST API + worker queue	giae_api/	giae serve / giae worker

---

Documentation

Topic	Link
Quickstart (5 minutes)	docs/quickstart.md
CLI reference	docs/cli.md
Python library API	docs/python_api.md
REST API reference	docs/rest_api.md
Architecture & confidence model	docs/architecture.md
Benchmarks (vs Bakta)	docs/benchmarks.md
Deployment (Docker, scaling)	docs/deployment.md
Extending GIAE (plugins)	docs/extending.md
Roadmap	docs/roadmap.md
Contributing	CONTRIBUTING.md
Hosted docs	Ayo-Cyber.github.io/GIAE/

---

Roadmap

GIAE is an evolving platform. The next horizon:

• Foldseek / AlphaFold structural homology — STRUCTURAL_HOMOLOGY evidence to crack PhiX174-class cases where sequence-based methods plateau
• Bacterial genome scaling — currently validated on phages; next target is 4–6 Mb bacterial genomes
• Translational coupling detection — recover compact-phage overlapping genes that lack canonical SD signals
• Comparative-genomics mode — diff two interpretations side-by-side
• Hosted SaaS — see PRODUCT_STRATEGY.md for the broader vision

Detailed roadmap →

---

Contributing

Contributions are welcome — issues, PRs, benchmark genomes, plugin ideas. Read CONTRIBUTING.md before opening a PR.

# Quickly run the test suite
pytest tests/ -q

# Run the Bakta head-to-head (requires Bakta installed)
.venv/bin/python post_assets/bakta_comparison.py

By participating you agree to the Code of Conduct. Security issues should follow SECURITY.md.

---

Citation

If GIAE contributes to your published work, please cite:

@software{giae_2026,
  author  = {{GIAE Contributors}},
  title   = {GIAE: Genome Interpretation and Annotation Engine},
  year    = {2026},
  version = {0.2.2},
  url     = {https://github.com/Ayo-Cyber/GIAE},
}

A formal application-note manuscript is in preparation (Phase 8).

---

License

MIT — use it, fork it, build on it. We only ask that you cite GIAE when it contributes to a publication.

---

_{Built by people who think genome annotation should be auditable.
Lambda 79.2% · T7 88.1% · 166 tests · 0 dark patterns}