How artificial intelligence can accelerate drug development

Artificial intelligence — the ultimate optimization engine — is meeting one of its biggest challenges: untangling the messy, slow and expensive work of drug development.

Why it matters: Even as computing power has gotten faster and cheaper, drugs remain slow and costly to develop, in part because of the sheer work in selecting a candidate and getting it across the finish line.

  • AI — with its ability to rapidly identify patterns in galaxies of data — can provide a vital shortcut.

What they’re saying: “What you’re seeing [with AI] is a platform for a new generation of drugs, biologics, life extension, all of which is being built at a rate that is impossible to describe,” says Eric Schmidt, former CEO of Google and the co-author of the new book “The Age of AI.”

The big picture: Drug development is a great business — if you don’t mind repeated, expensive failure.

  • The process of discovering and developing a new drug can take over a decade and costs $2.8 billion on average — and even then, 9 out of 10 therapeutic molecules fail Phase II clinical trials and regulatory approval.
  • The possible failure points are many — identifying a candidate drug from the more than 10⁶⁰ atomic configurations that exist in chemical space, optimizing it for delivery, and testing it in animals and humans to see if it is both safe and effective — and they all add to the overall cost of drugs and health care.
  • “Imagine you’re building 10 skyscrapers, and you can guarantee that nine will crumble,” says Isaac Bentwich, CEO of the new AI drug discovery startup Quris. “But you have no idea which ones will fall, so all you can do is build them and charge a higher rent on the one that keeps standing.”
  • “That’s the problem we’re trying to solve.”

How it works: AI can offer a boost at nearly every point of the drug development cycle, evangelists argue.

  • Exscientia’s “Centaur Chemist” AI platform computationally sorts through and compares millions of potential small molecules, looking for a handful to synthesize, test and optimize in the lab before selecting a candidate for clinical trials — all of which enabled the company to help get a cancer drug into trials in just eight months, compared to a more standard four to five years.
  • Quris is working to speed the trial process by testing drugs on miniaturized organs and tissues on a chip that “represent the full genomic diversity of the potential patient population,” notes Bentwich, which in turn generates data that can help train its AI platform to predict the clinical safety and efficacy of novel drugs.
  • Lantern Pharma is partnering with digital health care company Deep Lens to use AI to match the right kind of novel molecule with the right patient profile for clinical trials for accelerated clinical trials.
  • That AI-driven approach “can rescue hundreds of millions of dollars in prior drug development costs by ensuring it’s being tested on a very specific patient platform,” says Panna Sharma, Lantern Pharma’s CEO.

By the numbers: The AI market for pharmaceuticals rose from $200 million in 2015 to $700 million in 2018, and it’s projected to grow to $5 billion by 2024, while AI-related job listings in the drug industry have tripled over the past two years.

The catch: Even as AI becomes more powerful, the data sets for drug development can involve millions of compounds, which can exceed the capabilities of current machine learning tools.

  • The ultimate ability of AI to change the fundamentals around an industry as enormous, as expensive, and as regulated as drug development is “yet to be proven,” Paul Nioi, senior director of research at Alnylam Pharmaceuticals, told Genetic Engineering and Biotechnology News.

The bottom line: Sharma argues that will change — over the past 20 years, technology has “smashed the product development costs in everything except drug development,” he says.

  • “And over the next 20 years, it’s going to totally change this industry.”

Go deeper: How AI could revolutionize biology — and vice versa