LONDON – Owkin Inc. added a further $18 million to its series A, bringing the amount raised in the round to $70 million and equipping the company to push forward with its federated learning approach to applying artificial intelligence (AI) to the analysis of health data.

Closure of the round follows several demonstrations that federated learning can overcome ethical and privacy obstacles that have stood in the way of patient data being pooled for research.

Rather than gathering data in a single repository, distributed datasets are interrogated in situ, using algorithms downloaded from Owkin. Updated algorithms are then transferred back to the company.

Hospitals always remain in full control of their data and can benefit from the insights that come from training AI tools on big datasets.

The technique of federated learning first was proposed by Google in 2017 as a way of assuaging concerns about how the tech company handles user data.

Anna Huyghues-Despointes, head of strategy at New York-based Owkin, said the series A closing is a sign that the technique now is ready for widespread application in health. “I wouldn’t have said that two to three years ago. Then Google and big tech companies were comfortable with it, but others weren’t. We’ve moved on from when federated learning was seen as too cutting edge. In the next three to five years, it will be applied at scale,” she told BioWorld.

A potent example is the COVID-19 open AI consortium Owkin has launched to support collaborative research and accelerate development of therapies for patients with pre-existing cardiovascular conditions who contract the virus.

The Franco-American AI specialist is working with Capacity, a registry that currently includes histories of 4,222 COVID-19 patients treated at 62 hospitals, mostly in Europe.

“We are focusing on specific questions to build understanding of the complications [suffered] by cardiovascular patients affected by COVID-19,” Huyghues-Despointes said. Owkin has coordinated key opinion leaders in each of the institutions to define the questions and align the protocols for data preparation.

Having access to such a large number of patients will give a clearer picture of how different treatments affect outcomes, make it possible to assess relative risk of complications and severe disease to triage patients, and increase understanding of the immune response to the virus. Owkin’s AI tools are able to interrogate different types of patient data, including X-rays, blood tests and response to different drugs.

The COVID-19 pandemic has underlined further the need for access to high-quality, standardized patient datasets to enable collaborative research. By removing the requirement to get ethical approvals to use data distributed around different hospitals, federated learning has made it possible to analyze a large volume of COVID-19 patient data in a short time, while excluding artifacts that could be introduced as a result of differences between centers in how data are acquired and handled.

The initial findings have been submitted to a journal, and Huyghues-Despointes said once published, all the data will be made open source. “The model is trained on representative datasets and will generalize to other populations,” she said.

Another proof of the federated learning pudding is the EU-funded MELLODDY project, which is training machine learning tools on the chemical libraries of 10 European pharma companies and six other technical partners. This is making it possible for each company to learn from the others’ proprietary data while retaining control of their own information at all times. The aim is to develop algorithms that predict which compounds will have the most potential in future drug discovery and development work.

Owkin has received €2.8 million (US$3.2 million) for its part in the €18.4 million project.

“[Pharma] companies don’t want to share data, but they want to benefit from training on the model. This is a big proof point for pharma. Every pharma in MELLODDY now appreciates the [value of] federated learning,” said Huyghues-Despointes.

While MELLODDY is a publicly funded consortium, Owkin also has signed commercial deals with seven pharma companies. Here its AI and machine learning tools are applied to inform the design and analysis of clinical trials, biomarker discovery, patient stratification and the development of companion diagnostics.

Armed with the new funding, there now will be a push to sign up 15 more pharma companies. At the same time, Owkin will scale up the current network of 20 hospitals where it can send its algorithms – to 50 in Europe and 30 in the U.S. – to increase the volume of data it has to work with.

No Comments