Owkin taps Linux Foundation to open-source its AI learning software

Just a few months after publishing the results of a project wherein they convinced ten competing pharma companies to give the same AI access to their data sets, life sciences AI specialist Owkin is taking another step towards de-siloing drug development data: It's making its federated learning software Substra open source and placing it in the charge of the Linux foundation.

Substra is the software that powered the MELLODDY platform, a collaboration in which Amgen, Astellas, AstraZeneca, Bayer, Boehringer Ingelheim, GSK, Janssen, Merck, Novartis, and Servier all agreed to share drug discovery data – but not with each other directly, nor with Owkin. Instead, an AI learned from each of the company's datasets in turn, retaining no actual data but lots of aggregate impressions. The results of the collaboration were announced earlier this year, showing that the collaboratively trained AI outperformed single-partner models on multiple axes.

The now-open-source software also powers the Voice as a Biomarker of Health project, a $14 million NIH-supported research project in the US to see whether cancer and other diseases can be diagnosed by picking up subtle changes in a person’s voice patterns, and HealthChain, a project using similar methods as MELLODDY to enable collaboration between hospitals without compromising their proprietary data.

The Linux Foundation AI & Data Foundation will house the software, which will now be available for any researcher to use for similar collaborative data AI projects.

“Innovation thrives in collaboration, not in isolation – and the open sourcing of Substra is a landmark moment in the use of collaborative AI in medical research," Ibrahim Haddad, general manager of the LF AI & Data Foundation, said in a statement. "Researchers can now leverage privacy-preserving and secure federated learning software to drive cutting-edge collaborative medical research. Open source is undoubtedly the future of AI research.”

Owkin is simultaneously releasing two other open-source projects: FLamby, a federated learning model-ready dataset that researchers can use to develop models before training them on other data sets, and SecureFedYJ, a tool for normalising real-world healthcare data while maintaining data privacy.

"The future of medical research is collaborative," Owkin Chief Data Officer Mathieu Galtier said in a statement. "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. Owkin is committed to unlocking the vast potential held within patient data by developing technologies that overcome the privacy and security challenges that until now impacted research.”

A number of companies have sprung up offering AI models to help pharma companies with drug discovery over the past few years. Owkin, in particular, has bagged some high-profile, high-price tag development deals with the likes of Bristol-Myers Squibb and Sanofi.