v0.3.0 — Re-ranker, evaluation pipeline, annotate workflow, UI overhaul

Release v0.3.0 — Re-ranker, evaluation pipeline, annotate workflow, UI overhaul

Features

• Re-ranker neural model: temporal holdout training pipeline with LightGBM, feature engineering (alignments, taxonomy), and scoring configs
• CAFA evaluation pipeline: automated evaluation with multiple metrics (Fmax, Smin, AUPR)
• Annotate workflow: end-to-end functional annotation from FASTA upload to GO term prediction
• Scoring engine: configurable scoring configs with evidence weights
• Connection pool, DLQ, structured logging, health probes, stale job reaper
• Full i18n: 5 locales (EN/ES/DE/PT/ZH) via next-intl
• Frontend overhaul: scoring config UI, support page, evaluation views

Tests

• Coverage expanded from 65% to 88% (283 → 831 tests)

Docs

• ADRs, operational runbook, re-ranker design spec
• Full Sphinx documentation update

CI

• Bump GitHub Actions to v6 (checkout, setup-python)
• Fix all ruff, flake8, and mypy lint errors

v0.2.0 — Scoring, Evaluation & i18n

What's new in v0.2.0

Core

• Scoring engine — ScoringConfig model with linear and evidence_weighted formulas; five built-in presets (embedding_only, embedding_plus_evidence, alignment_weighted, composite, evidence_primary)
• CAFA evaluation pipeline — NK/LK/PK ground-truth delta, Fmax / AUC-PR metrics endpoint, EvaluationSet and EvaluationResult ORM models

API

• POST /scoring/configs — full CRUD for scoring configurations
• POST /scoring/configs/presets — seed built-in presets (idempotent)
• GET /scoring/prediction-sets/{id}/score.tsv — stream scored predictions TSV
• GET /scoring/prediction-sets/{id}/metrics — compute CAFA Fmax / AUC-PR
• Extended annotation endpoints: evaluation sets, per-namespace metric views

Frontend

• Scoring config UI — create, view and delete configs; apply to prediction sets
• Full i18n — EN / ES / DE / PT / ZH via next-intl
• Human-readable labels in prediction set list
• Support page

Infrastructure

• Docker production deployment via ghcr.io/frapercan/protea:latest
• CI: build and push images on every push to main
• manage.sh: process-group kill fix (setsid)

Tests & Docs

• New test suites: scoring, metrics, evaluation, routers
• Architecture docs: evaluation pipeline, updated operations and core reference
• NumPy 2.0 compatibility fix (np.trapezoid)

v0.1.0 — Initial Release

PROTEA v0.1.0

First official release of the PROTEA platform.

Features

• Protein ingestion from UniProt (REST API, cursor-based pagination)
• UniProt metadata fetching
• GO ontology loading (OBO format)
• GO annotation loading (GAF format — GOA, QuickGO)
• Protein embedding computation (ESM2)
• GO term prediction via KNN transfer
• CAFA evaluation pipeline
• FastAPI REST API with 21 endpoints
• Next.js frontend
• RabbitMQ job queue with 7 routing queues
• Full CI/CD with GitHub Actions
• Docker support (docker compose up)

Docker

docker compose up

Images published at ghcr.io/frapercan/protea.