Three ways to help bacterial-defense research — without setting foot in a lab.

In May 2026, the Institut Pasteur published an atlas of 2.39 million bacterial proteins likely involved in antiviral defense, grouped into 478,206 families. The GeneCLRDF AI model that identified them is 99% accurate, but 85% of these families had never been associated with immunity before. Only 12 systems have been experimentally validated to date.

Pasteur publishes the data. The human layer to explore, sort and derive medicines from it is missing. That's what Bactaegion builds.

Organize the exploration of the Pasteur atlas.

The problem: 478,206 candidate families is unmanageable for a single human. No laboratory has time to test each in in vitro validation. A human layer is needed to prioritize.

What you do here: you learn to recognize the patterns researchers use — your eye calibrates on 12 piano-roll scenarios, 6 proteins to annotate, dossiers to read. Tutorial on already-resolved cases, no scientific pressure.

What it changes: eventually (J3 of the roadmap, see /en/roadmap/), you'll flag real orphan families from the GeneCLRDF dataset. Pasteur or any defense lab will receive a prioritized list of candidates to test — direct time savings on validation cycles.

Produce pharmacological hits in open science.

The problem: bacterial defenses lead to real human medicines. Sofosbuvir (anti-hepatitis C) is an analog of bacterial viperin chemistry. Disarm Therapeutics, which developed SARM1 inhibitors inspired by the Thoeris system, was acquired by Lilly for $1.3 billion in 2021. But the industry selects what it exploits — many targets remain on the shelf, or get patented restrictively before a community can work on them.

What you do here: you peer-review sourced therapeutic hypotheses on 3 criteria (reproducibility, parsimony, rigor). If you're a scientist and want to formulate your own hypothesis, you can plug in your own LLM engine (Claude, Gemini, local Ollama) — your key is AES-GCM encrypted in your browser, never transmitted to Bactaegion.

Priority targets identified by translational audit: Viperins (nucleotide analogs in the Sofosbuvir family), CBASS (cGAS-STING axis in immuno-oncology), Pycsar (cyclic pyrimidines), Schlafen (tRNA modulators), RADAR (ADAR1 in interferonopathies), Thoeris (SARM1 in neurodegeneration).

What it changes: all contributive hypotheses are released under CC0. Eventually, the clinical handover is done via non-commercial sponsors (DNDi, GARDP, M4K Pharma model) — not via private capture. Bactaegion goes up to Hit-to-Lead, the rest belongs to structures with pharmaceutical regulatory infrastructure.

Enable confidential laboratories via Federated Learning.

The problem: many laboratories (hospitals, biotechs) hold clinical or environmental genomes under NDA. They cannot share raw data for legal (GDPR) or industrial (pharmaceutical secrecy) reasons. But they could enrich the GeneCLRDF model if a method existed.

Current state: not started. Target identified in the roadmap (milestone J6) via the DAP/Daphne protocol (Cloudflare, open source). Each lab would train locally on its data, sending only encrypted gradient fragments. Depends on recruiting laboratory partners — as much strategic work as technical work.

We display this objective honestly: it's the vision, not today's reality. See /en/roadmap/ for the real timeline.

• No XP, leaderboard, "level" badges, streaks, guilt-tripping push notifications.
• No loot boxes, opaque randomness, artificial scarcity.
• No speculative crypto-token (use value doesn't run through a financial asset).
• No required sign-up. No PII collection. Your contributions belong to you.