[Reporter’s View] Finding candidates with AI is no longer enough

A drug candidate discovered by artificial intelligence has finally reached the Phase III clinical trial stage. Global AI drug discovery company Insilico Medicine dosed the first patient in a Phase III trial of rentosertib, its treatment for idiopathic pulmonary fibrosis, on the 10th. A candidate in which AI was involved from target discovery through molecular design has now reached the stage of testing its efficacy and safety in a large patient population.

Rentosertib’s entry into a Phase III trial is a symbolic milestone in the field of AI drug discovery. Until now, AI achievements have largely been described in terms of how many compounds were analyzed or how much time could be saved in the candidate discovery stage. The field has now entered a stage where the question is whether AI-discovered candidates actually work in patients.

Coincidentally, this year marks a decade since Korean Go champion Sedol Lee l faced AlphaGo in their historic match. AI, which demonstrated its potential on the Go board 10 years ago, has now advanced to the point where it can identify drug candidates that progress all the way to Phase III trials.

The use of AI across the pharmaceutical and biotech industry has also expanded rapidly. Its applications now extend beyond drug target identification and candidate design to clinical trial design, manufacturing process optimization, and sales and marketing analytics. Major Korean pharmaceutical and biotech companies are also building their own AI platforms or conducting joint research with specialized AI companies.

The change brought by the adoption of AI has been particularly evident in the early stages of drug development. By rapidly exploring vast chemical spaces that would be difficult for humans to review individually and narrowing them down to promising candidates, AI has demonstrated the potential to reduce trial and error and shorten timelines in early drug discovery. Rentosertib is regarded as a prime example of a candidate identified far more quickly than would typically be possible with conventional drug discovery approaches.

The issue is that the yardsticks used to assess AI drug discovery have not kept pace with the technology’s development. Achievements such as building an AI platform or identifying a candidate within a matter of months still receive considerable attention. Far fewer examples, however, show how well those candidates survive clinical development or how much AI can reduce failure rates and development costs compared with conventional approaches.

Of course, the ability to use AI to identify drug candidates and shorten development timelines was itself a clear achievement 10 years ago—or even just 5 years ago. Simply demonstrating that AI could be applied to real-world drug development was meaningful enough.

But the situation has changed. Now that AI-discovered candidates are progressing beyond discovery and into Phase III trials, the standards by which the technology is judged should also move up a level. More important than how quickly or how many candidates AI can identify is how many of those candidates ultimately succeed in clinical trials, and how much AI can reduce overall development time, costs, and the risk of failure. Unless it can prove those benefits, AI may remain a powerful tool for making candidate discovery more efficient rather than an innovation that fundamentally transforms drug development.

In that sense, Rentosertib’s entry into Phase III marks not the success of AI drug discovery, but the beginning of its real test. Ten years after AlphaGo, what AI drug discovery needs now is not another platform or another tally of candidates. It needs evidence that AI can actually improve the odds of successfully developing a new drug.