Machine learning drives progress in fibrosis treatment

Alex Zhavoronkov, a biotechnologist and computer scientist, founded Insilico Medicine in 2014 because he was confident that artificial intelligence (AI) held the key to speeding up and lowering the cost of drug discovery and development.

Over the years, he and his team have developed multiple tools for identifying drug targets and developing therapies, but one question has always hung over their efforts: Could a drug discovered and designed by AI actually succeed in the clinic?

Now, Zhavoronkov and Insilico have moved one step closer to reaching that ultimate goal — successful completion of their Phase 2a clinical trial for rentosertib, a novel drug that Insilico’s software designed to treat idiopathic pulmonary fibrosis (IPF) (1).

“This trial represents one of the first real-world validations of a fully AI-driven drug discovery process,” Zhavoronkov said.

Alex Zhavoronkov and his colleagues at Insilico Medicine want to use AI to accelerate drug discovery and development.

Credit: Alex Zhavoronkov

At the outset, Zhavoronkov and his team were interested in developing general treatments for fibrotic diseases, which can cause life-threatening accumulation of scar tissue not just in the lungs, but also in the kidneys, liver, and other organs.

Previously, the researchers used PandaOmics, a machine learning tool that combs through transcriptomic and proteomic data, to identify the protein Traf2- and Nck-interacting kinase (TNIK) as a novel therapeutic target involved in fibrosis across multiple organs (2).

In that same study, the Insilico team then used Chemistry42, an AI platform that generates novel drugs using filters for effectiveness, safety, and feasibility, to design rentosertib. The small molecule reduced inflammation and scarring in multiple preclinical models of IPF and proved safe for healthy volunteers in a Phase 1 trial. “From target selection to IND-enabling studies, the process took approximately 18 months,” Zhavoronkov said.

In the current study, Zhavoronkov and his colleagues carried out a Phase 2a trial in China, where they enrolled 71 people with IPF. The researchers included both people who were not taking any other disease-modifying medications and people who were already taking nintedanib or pirfenidone, the two currently available treatments for IPF.

Participants took daily doses of rentosertib for 12 weeks, with side effects occurring infrequently and most often in those also taking nintedanib. Furthermore, patients who took 60mg of rentosertib tended to improve on a breathing test at the end of the 12 weeks, while patients who received a placebo worsened.

“This positive signal in a placebo-controlled setting was encouraging and definitely supports further clinical development,” Zhavoronkov said.

Simon Cruwys, a pharmacologist with TherapeutAix who was not involved in the study, was unsure how much success the Insilico team would have in a larger Phase 2b or Phase 3 trial based on what he has seen so far.

Cruwys noted that, despite the novel AI platforms they used in developing rentosertib, Zhavoronkov’s team followed a standard industry playbook for IPF, including the molecular pathways and datasets they mined for therapeutic targets and the preclinical models they used to test their drug. “That industry standard has failed to produce a new therapeutic in 14 years,” Cruwys said. During that time, he noted that most attempts have failed in later clinical trials, and Insilico has yet to clear that hurdle.

This trial represents one of the first real-world validations of a fully AI-driven drug discovery process.
– Alex Zhavoronkov, Insilico Medicine

Cruwys was cautiously optimistic about TNIK as a novel drug target for IPF because the protein is involved in multiple fibrotic pathways, but he would need to have more information about how rentosertib interacts with nintedanib and pirfenidone to know if it is the right molecule for the job. While Zhavoronkov and his team found that their drug improved patients’ breathing capacity, it did not provide any additional benefit when combined with the two already available therapeutics. Before moving on to a larger clinical trial, Cruwys said, “I’d want to explain why they’re not seeing additivity.”

Cruwys said that an IPF drug that cannot provide an added benefit when combined with the current standard of care would struggle to break through and achieve market viability. He explained that, although nintedanib and pirfenidone are not ideal because of their side effects and limited benefits, they are about to become cheaper as they move off-patent. Thus, they would be difficult to displace.

Zhavoronkov and Insilico plan to begin a Phase 3 trial of rentosertib later this year. While success would further validate their AI-driven approach, Zhavoronkov believes their progress so far already demonstrates the technology’s potential. “Rentosertib is a case study for the potential of AI to help reduce timelines, discover novel targets, and generate compounds with real clinical benefits,” he said.

References