Collaborations Pharmaceuticals Featured in New Netflix Documentary, Highlights Risks of AI

The small team at Raleigh-based startup, Collaborations Pharmaceuticals, spends the bulk of its time looking for safe and effective molecules that can be developed into drugs to treat rare and infectious diseases.

Their work requires the use of artificial intelligence (AI) – principally machine learning software – to mine vast amounts of publicly available scientific data.

But, about two years ago, the Collaborations paradigm was briefly turned on its head. That’s when the company’s founder and CEO, Sean Ekins, was asked to speak at an international conference in Switzerland on – of all things – chemical and biological weapons. “They probably pulled me in because of all the work I’ve done in the past on toxicology modeling,” he said.

The premise: what would happen if Collaborations’ MegaSyn AI drug discovery platform was used for more nefarious purposes? “They were mainly interested in the misuse of technology,” Ekins pointed out. “I’m running a small drug discovery company, so the misuse of technology is the last thing on my mind.”

Ekins and Fabio Urbina, another Collaborations scientist, decided to tweak their drug discovery platform to see what results they might produce. MegaSyn – which is designed to find the most specific and least toxic molecules – was reprogramed to generate VX, one of the deadliest and fastest-acting chemical warfare nerve agents.

It wasn’t much of a stretch. Collaborations had previously used MegaSyn to explore non-toxic molecules to treat neurodegenerative conditions such as Alzheimer’s disease that have the same molecular target as VX.  

‘Overwhelming, quite surprising, and a little surreal’ 

So, all Ekins and Urbina had to do was revise the end goal toxicity and then set a threshold to only generate molecules as lethal as VX. “For me, it was trying to see if the technology could do it,” Ekins explained.

The program ran overnight and “unraveled in a way that was totally unpredictable.” Within six hours, a 2015 vintage Mac computer turned out close to 40,000 different substances potentially as deadly as VX. Some might be more dangerous, and thousands weren’t listed in any public database. 

“It did a remarkable job of coming up with designs for potentially toxic molecules,” he observed.

No further testing was done to determine the viability of any of the substances as potential biological weapons. But the Ekins/Urbina study highlighted the possibility. And it demonstrated the ease with which AI could rapidly generate huge amounts of data that might speed up the development process.

“Looking at the information on the computer screen was overwhelming, quite surprising, and a little surreal,” Ekins explained. “It struck me how easy it was to pull up this sort of data from public sources with off-the-shelf technology.”

When Ekins discussed his findings at the Swiss Spiez CONVERGENCE conference – held every two years to consider new trends in biological and chemical research that could threaten national security – “you could almost hear a pin drop,” he said. “It created an overwhelming amount of discussion.”

A peer-reviewed article published in the March 2022 edition of the journal Nature Machine Intelligence (“Dual use of artificial-intelligence-powered drug discovery”), reached an exponentially larger audience. Ekins’ life hasn’t been the same since.

The study went viral and generated a huge amount of attention. Ekins met with the White House Office of Science and Technology Policy, the CIA, the Department of Defense and the State Department. And that was just in the United States.

The media jumped on board, with articles appearing in well-regarded scientific and business publications. Ekins also discusses his findings in a new Netflix documentary focused on the use of AI in the military that aired this week, trailer here.

He said the experience has raised his awareness and changed the way he looks at AI.

“While we assume everyone will use scientific data for positive purposes, we know that’s not true,” Ekins pointed out. “We already have the weight of the world on our shoulders as we try to come up with drugs to treat really horrific diseases, and now we must think about how to prevent the misuse of technologies we’ve been trying to use for good.”

He suggests more discussion within the pharmaceutical and biotechnology industries on how to regulate and administer AI applications for drug discovery. That could mean rethinking what data and methods are made publicly available, more closely tracking open-source datasets, and putting in place ethical oversight committees for AI. Education and ethical training also are important.

“There will be more focus on regulation, which is a positive, but we don’t know how that will affect our industry,” Ekins said. “AI has a lot of value, but there also is a dark side. And if the technology backfires, it will have a tremendously negative impact on all of us.”

About Collaborations Pharma

Collaborations Pharmaceuticals, which Ekins founded in 2015, partners with the academic community, government agencies, foundations and other companies to develop drugs that treat diseases that otherwise might be neglected.

It uses machine learning software to analyze publicly available data, organizes and integrates that information, and then builds models to find the most promising molecules to treat specific diseases.

The company’s goal is to advance new therapies from early stage development to clinical trials. It’s lead project is an enzyme replacement therapy for Batten disease, an ultra-rare and fatal inherited nervous system disorder that most often begins in childhood.

Collaborations employs just over a dozen people and is located on the Centennial Campus at North Carolina State University.

Most of the company’s operating capital comes from grants to promote the development of drugs that are often overlooked by large pharma companies. Collaborations has received close to $17 million in federal funding since 2016.