Over the past three years, 10 companies in what is arguably one of the most competitive industries in the world have been sharing data in unusual – and once unthinkable – ways.

An EU-funded project using artificial intelligence (AI) to boost drug discovery has just been completed, and it has shown its potential to transform the pharmaceutical industry, according to Owkin, the Franco-American unicorn behind it.

The objective is to accelerate the discovery and development of new treatments, which generally require hundreds of millions of euros of research and take more than 10 years to reach the market.

Major pharmaceutical companies such as GSK, AstraZeneca and Novartis have collaborated on the program, called MELLODDY, which allowed Owkin’s machine learning models to train on their confidential research without drugmakers having to worry that this valuable data leaves their own servers.

“Instead of collecting the data, we let the model travel from one pharma to another,” Mathieu Galtier, Owkin’s Chief Data Officer, told Euronews Next in an interview.

“It’s a kind of symphony. That’s why we liked the melody of the world; we orchestrate how each model is going to learn a little about this pharma, then about another”.

And there was a lot to learn.

The MELLODDY platform has trained AI models on billions of industrial experimental data points, documenting the behavior of more than 20 million small chemical molecules in more than 40,000 biological tests.

A “spotlight” in drug discovery

Owkin’s goal was to harness all of this data to make his machine learning tools smarter and better able to predict how a certain compound will react or bind to certain proteins – and therefore which compound will be suitable for a certain drug. or therapeutic target.

This is where AI has the potential to radically disrupt the pharmaceutical industry and the way it discovers new treatments, says Galtier.

“In the past, we identified a target, saying that it was the protein we wanted to activate or inhibit. And then we were trying all the different compounds in the world, and we were doing random things,” he said.

“Now with artificial intelligence, the goal is to move towards algorithms that suggest new molecules, algorithms that will enlighten you in this approach”.

Owkin was founded in 2016 and last year achieved “unicorn” status – when a startup is valued at over $1 billion (976 million euros) – following a $160 investment. million euros from the French drugmaker Sanofi.

Owkin developed his AI tools by distributing them across medical and research centers, working with local servers to access the data on which he trains his algorithms.

Since Owkin does not collect the data on a central server, the MELLODDY platform uses blockchain technology, with a ledger distributed to all contributing pharmaceutical partners, to keep track of its activities.

The MELLODDY project was funded by the Innovative Medicines Initiative (IMI), a partnership between the European Union and the European Federation of Pharmaceutical Industries and Associations (EFPIA).

Project officials say the initiative not only demonstrated that Big Pharma could cooperate, it was also more effective by sharing data in this way.

The trial proved that the “collaborative” model used for the MELLODDY project was on average 4% better than stand-alone AI models from drugmakers at classifying molecules as pharmacologically or toxicologically active or not.

The data sharing also increased the model’s ability to make reliable predictions when examining new types of molecules by 10% – known as the “domain of applicability”.

“It’s better to collaborate than to go it alone, and that’s the main message we took from the whole project,” Galtier said.

Pharmaceutical lobby EFPIA agreed, saying the pilot could convince drugmakers to collaborate and open up their data to boost drug discovery.

“MELLODDY has demonstrated the feasibility of secure collaborative modeling without jeopardizing intellectual property,” an EFPIA spokesperson told Euronews Next, adding that the project had “enabled each partner to achieve better predictive models for the majority of their discovery tests”.

While a 4% increase in predictive power might not seem so impressive to a layman, it’s an average based on data from tens of thousands of drug discovery experiments – and that could mask a much larger increase for some of these experiments, Galtier said. .

“What’s going to have an impact are the few drugs where the improvement is greater than 20%,” he said.

Suspicious pharmaceutical

It was still very difficult to involve the pharmaceutical companies in the early stages of the project.

Not only did Owkin have to convince scientists to share their data, but he also had to assure legal teams and corporate security experts that the data would be safe.

This may have had an impact on the type of data they initially agreed to share, some of which may not have been the most relevant or insightful.

“If you put yourself in the shoes of a pharmaceutical company… You’ll start with something that’s safe, like the good old tests, the experiments they did 10 years ago,” Galtier said.

However, as the project has progressed, larger volumes of data have been shared, including from so-called “active experiments” where drugmakers are currently researching.

But Galtier knows nothing of the underlying data they provided. None of the other operators on the platform either – that’s the whole premise it was built on.

“I’m sure if Owkin or if I had been given the right to see the data, the pharmaceutical industry wouldn’t have signed on, they would have said no,” Galtier said.

The AI ​​model simply moves from one set of pharmaceutical data to another and learns on the spot; it retains statistical information about the data, such as averages, but no information about the underlying data on which it is trained, such as the actual chemical compounds studied.

“We made sure of that. We had academic partners walking around and trying to attack the models, extract information… We made sure that was not possible. It was one of the prerequisites for pharmaceutical companies,” Galtier said.

Owkin says the successful deployment of MELLODDY has reassured Big Pharma that data can be shared in a safe way, and he now plans to create more specialized “channels” or “consortia” for drugmakers wanting to share data using the same federated learning technology.

The shared data would go beyond drugs and small molecules to include protein design, antibodies and patient data – all potential treasure troves of information to help identify new treatments for cancer, diabetes or diseases. neurodegenerative diseases such as Alzheimer’s disease.

Galtier said addressing drugmakers’ privacy concerns had opened up a new space for collaboration, which he called “coo-petition”: no competition, no collaboration, but something in between.

“We are showing that it works. It’s a first step, but it’s a very good first step”.