FL-CHEMSAFE


Federated Chemical Safety Assessment (FL-CHEMSAFE) is an initiative by Dr Nicoleta Spînu.

Frustrated by the constant trade-offs between working with limited, siloed and fragmented data, the need to adapt (novel) methodologies to data size and availability, ending up solving one task within a problem space while biology is much more complex, FL-CHEMSAFE was set out to enable a more integrated, holistic and adaptive approach to chemical safety assessment.

Our goal is to facilitate knowledge transfer and privacy-preserved collaborations for comprehensive and better informed decision-making, contributing to the real-world adoption of New Approach Methodologies for Next Generation Risk Assessment.

Federated skin sensitisation QSAR use case

Use Case Example: Federated skin sensitisation logistic regression QSAR demonstrating how organisations with different compound distributions can collaborate to create a more balanced and accurate global model while keeping their data private.

Join us

We are currently designing the prototype. Get in touch and help shape its direction.

New to federated approaches?

How can federated approaches help you and your team?

Here are some examples:

What can we do for your team now?

Frequently Asked Questions

Yes, privacy and security are at the core of federated approaches. By using federated pipelines, your proprietary or sensitive data never leaves your premises. Also, Flower is ISO 27001 certified.

This is part of the design of the federated pipeline, where parties have to discuss and agree on the type of information and parameters to be co-trained or shared. There are various federated methods, and Flower facilitates their implementation and allows for personalised collaborations!

Yes, our simulated use case on mutagenicity, where one organisation had 80% mutagenic compounds and the other organisation had 20% mutagenic compounds, showed that the federated global model outperformed both local models and had a balanced performance overall. Thus, the global model was able to differentiate between mutagenic and non-mutagenic compared to a local model.

There are various approaches. The most used one is cyclical, where the local models are co-trained simultaneously in multiple rounds. We are happy to run live workshops. Get in touch!

Researchers, regulatory experts, industry professionals, raw material suppliers, chemical manufacturers, consultancy, in vitro testing providers. FL-CHEMSAFE can help sharing knowledge across departments and stakeholders, flexible and scalable.

Access proprietary knowledge via model training and analytics without exposing your private data.

Supported by

Oral Presentation QSAR2025 Award 🏆

Aligned with the UN Sustainable Development Goals