France/US-headquartered AI biotech company Owkin is open sourcing the Artificial Intelligence (AI) software behind Melloddy and Substra.
Big fans of ‘open science’, Owkin hopes to help universities, hospitals and pharmaceutical companies to benefit from its privacy-preserving, secure and collaborative AI technologies.
Owkin is open sourcing Substra, its Federated Learning (FL) software, to allow researchers and developers to collaboratively train ML models without the data leaving its source.
The team has set its sights on overcoming data privacy and security barriers.
The move will enable users to use an AI technology that has already proven its ability to improve the performance of ML models.
Substra is hosted by the Linux Foundation – the LF AI & Data Foundation provides a neutral home for Substra based on open governance principles. Substra underpinned the Melloddy, an AI drug discovery collaboration uniting 10 pharmaceutical companies that demonstrated that collaborating in AI for drug discovery is possible at an industrial scale.
We can also note here that Substra is also powering the HealthChain consortium, a project enabling hospitals to develop collaborative AI models on diseases without the data leaving hospital firewalls. It is also being used in a landmark project to establish the human voice as a routine biomarker used to diagnose and treat diseases.
Also released is SecureFedYJ – a solution to help normalise real-world healthcare data distributed across multiple data centres in a federated manner, without compromising on data privacy or security.
“The future of medical research is collaborative. By open sourcing Substra and releasing two landmark federated innovations to researchers, we hope to unleash a wave of collaborative research that will spur on the development of the next generation of treatments,” said Mathieu Galtier, chief data officer at Owkin.
Owkin is also releasing FLamby as the world’s largest open-source FL-ready dataset designed to help researchers conduct initial experiments on different data modalities and find the most effective approaches before deploying models on real data.
FLamby is intended to help build a collaborative community that accelerates the use of FL in healthcare. FLamby is available on GitHub.