Back to blog

How do you kill a failing drug before it costs billions?

Thắng Đoàn
Thắng Đoàn

Novartis expected del-desiran to earn $5 billion a year. Instead, a late-stage miss wiped $30 billion off its market value. An AI start-up had called that failure in advance.

The most expensive guess in medicine

Per Reuters reporting, the industry spends about $140 billion a year on human clinical testing, yet only around 12% of drug candidates win approval. Reuters says that success rate has changed little in decades. Every late-stage failure burns years of work and investor confidence overnight. The core problem: you cannot see the outcome until you have already paid for the trial.

  • Novartis shares fell 11% the day del-desiran missed its goal.
  • Its CEO had projected $5 billion in annual peak sales.
  • Reuters puts the approval rate near 12%, flat for decades.

Simulated patients before real ones

In July, AI start-up BioinvestGPT ran a simulated trial of del-desiran and predicted an insignificant clinical benefit, Reuters reports. It and other AI firms now work with drugmakers on virtual trials. The idea is to test a drug on digital patients first, then spend real money only on likely winners. Reuters also reports simulations are used to decide which programs advance and to assess acquisition targets.

  • Simulations run before the human trial, not after the failure.
  • Use cases go beyond trials: program funding and deal valuations.

Discovery is no longer the bottleneck

TechCrunch reports Enveda raised $311 million at a $2 billion valuation, led by Catalio Capital Management, to push AI-discovered, nature-derived drugs into human trials. That valuation doubled in 12 months. But TechCrunch notes AI has not yet produced any FDA-approved drug. Capital is now chasing candidates that survive trials, not just candidates that look good on paper.

  • Enveda is testing several drugs in patients, per TechCrunch.
  • The round was led by Catalio with participation from Iconiq.

What this does not solve

No AI company has an approved drug yet, per TechCrunch. A correct prediction announced after a public failure is easy to claim; Reuters reports no track record of simulations run before successful trials. Virtual patients can only model the biology their training data contains, so genuine surprises can still sink a program. And if pharma trusts simulations too much, promising drugs may be killed before anyone tests them.

Which of your expensive long-cycle bets could be stress-tested by a simulation before you commit more capital?
If a model predicted your flagship project would fail, like BioinvestGPT did with del-desiran, would you stop it?

List your three costliest ongoing projects and mark which one a cheap simulation could invalidate this quarter.

Sources

The Star (Reuters): AI companies run virtual drug trials

Yahoo Finance (Reuters): AI companies run virtual drug trials

TechCrunch: Enveda secures $311M

KSL (Reuters): AI companies run virtual drug trials

Share:

Recommended for you

Enjoyed this article?

Subscribe for new articles. No spam. Unsubscribe anytime.

By subscribing you agree to receive the newsletter. No spam, and you can unsubscribe anytime.