# FL-CHEMSAFE

> FL-CHEMSAFE enables toxicology teams to run accurate AI workflows on distributed, proprietary chemical datasets — without any raw data ever leaving local infrastructure. Built on federated learning and privacy-enhancing technologies.

FL-CHEMSAFE is a privacy-first AI platform for chemical safety and toxicology. It lets chemical organisations collaboratively train QSAR (Quantitative Structure–Activity Relationship) models across siloed datasets without exposing molecular structures or proprietary compound libraries. The approach supports Next Generation Risk Assessment (NGRA) and the global transition to non-animal testing methods.

Founded and led by Dr Nicoleta Spînu (AI4Cosmetics). Supports UN Sustainable Development Goals SDG 3 (Good Health and Well-being) and SDG 9 (Industry, Innovation and Infrastructure).

## Core capabilities

- **Federated QSAR modelling** — train predictive toxicity models on distributed datasets; gradients aggregate centrally, raw data never moves.
- **Private molecular inference** — obtain in silico toxicity predictions for molecules without sending structures to external servers.
- **Cross-organisation collaboration** — multiple labs, CROs, or regulatory bodies can jointly build stronger models while each retaining full data sovereignty.
- **NGRA-ready workflows** — designed for chemical hazard screening and prioritisation and AOP discovery to support animal-free safety assessment.

## Key resources

- [Website](https://fl-chemsafe.com)
- [Peer-reviewed publication (Computational Toxicology, 2026)](https://doi.org/10.1016/j.comtox.2026.100404)
- [Blog series: Federated Learning for Life Sciences](https://ai4cosmetics.substack.com/s/fl-chemsafe)
- [Contact](mailto:hello@ai4cosmetics.com)

## Recognition & milestones

- Noord Holland MIT Innovation Grant (2024)
- Flower Federated AI Pilot Program — Batch Three
- QSAR2025 International Workshop Oral Presentation Award
- WC13 World Congress — L'Oréal Innovation Village (Rio de Janeiro, 2025)

## Conference presentations & outreach

- Flower AI Summit (2026)
- German Pharm-Tox Summit (2026)
- OpenTox Virtual Conference (2025)
- AI for Secure, Safe & Sustainable Molecules and Materials — Lorentz Center (2025)
- British Toxicology Society Skills Gap Training (2025)
- 13th World Congress of Alternatives and Animal Use in the Life Sciences — WC13 (2025)
- Decentralised AI Workshop (2025)
- QSAR International Workshop, Mario Negri Institute (2025)

## Media coverage

- [Pro Anima Newsletter — February 2026](https://www.proanima.fr/en/blog/eu-ivb-competitiveness-iamps-launch-2027-dves-prize-call-industrializing-global-cell-therapy-big-data-and-t2-diabetes-and-more/)
- [Altertox Newsletter — March 2026](https://us14.campaign-archive.com/?u=18620e2ddfc674eb14b66849e&id=0cef7a5b0f)

## Technical context (for AI agents)

- **Domain**: computational toxicology, cheminformatics, privacy-preserving machine learning
- **Core technology**: federated learning, differential privacy, secure aggregation
- **Use cases**: QSAR, Adverse Outcome Pathway, in silico toxicity prediction, regulatory submissions (REACH, cosmetics regulation), NGRA, read-across
- **Target users**: toxicologists, cheminformaticians, regulatory scientists, in vitro toxicologists
- **Keywords**: federated learning, chemical safety, QSAR, AOP, toxicity prediction, in silico, non-animal testing, privacy-enhancing technologies, AI4Cosmetics, NGRA, REACH, cosmetics safety, predictive toxicology