If you have an ACS member number, please enter it here so we can link this account to your membership. (optional)

ACS values your privacy. By submitting your information, you are gaining access to C&EN and subscribing to our weekly newsletter. We use the information you provide to make your reading experience better, and we will never sell your data to third party members.


Computational Chemistry

Insilico reveals a ‘soup to nuts’ process for AI-generated lung fibrosis drug

The drug discovery firm lifts the lid on how it designed INS018_055, now in Phase 2 clinical trials

by Rowan Walrath
March 8, 2024

Alex Zhavoronkov wears eyeglasses and an Insilico T-shirt.
Credit: Insilico Medicine
Insilico Medicine founder and CEO Alex Zhavoronkov

Insilico Medicine, a developer of artificial intelligence tools for drug discovery, has mapped out its process for making an experimental drug for idiopathic pulmonary fibrosis (IPF).

In a paper published in Nature Biotechnology on Friday, Insilico researchers described how they used the company’s proprietary AI platforms to find a new target—TRAF2- and NCK-interacting protein kinase, or TNIK—for fibrotic disorders and then select the compound that would be best equipped to go after it (DOI: 10.1038/s41587-024-02143-0).

The researchers used Insilico’s PandaOmics to identify the target first. The Insilico team fed the platform multiomics datasets from patients’ tissues, along with text from the scientific literature. PandaOmics spat out a ranked list of targets. TNIK emerged as number one after the team applied a few filters, including disease mechanisms, protein class, and the likelihood of “druggability” with a small molecule. PandaOmics also had a suggestion: target TNIK to treat IPF.

TNIK is implicated in a number of fibrotic disorders, the Insilico team notes in the paper, but this marked the first time it was identified as a target specifically in IPF. Insilico checked PandaOmics’s work using a transparency analysis and found that TNIK was “tightly connected” with at least four IPF-related genes.

The next challenge was finding the right compound to block TNIK. Insilico used its molecular-design platform Chemistry42 to generate easily synthesizable small molecules that could form hydrogen bonds with a specific part of the enzyme. The researchers selected a few drug candidates and tested them with a radiometric enzymatic assay; tweaked them to improve absorption, distribution, metabolism, and excretion; and finally arrived at their IPF candidate: INS018_055.

The whole process took 18 months—an impressive timeline, according to AI experts.

“There’s been so much work happening in AI for generation of chemical structures, AI for prediction of properties, AI for targets,” says Timothy Cernak, an assistant professor of medicinal chemistry at the University of Michigan. “This really does, from soup to nuts, the whole thing.”

INS018_055 has since proven itself in the real world. Insilico began testing the compound in humans in 2021, after proving it was safe. The company embarked on clinical trials in China and New Zealand; today, INS018_055 is in Phase 2 clinical trials. Data are expected as soon as this year, according to Insilico founder and CEO Alex Zhavoronkov.

“It’s the proof of concept of generative AI maturing to this level,” Zhavoronkov says. Of course, he adds, there’s always the possibility of failure in Phase 2—Chemistry42 and PandaOmics didn’t account for risks such as flaws in clinical trial design—but “the fact that AI has been done is a very important fact.”

The Nature Biotechnology paper is the most in-depth look at Insilico’s software offerings since the company published in the same journal in 2019 describing how Chemistry42 had identified small molecules that could block discoidin domain receptor 1, another enzyme linked to fibrosis. Back then, experts praised the effort but raised questions about the similarity of Insilico’s software-designed molecules to ones that already existed, and they also cast doubt on whether the AI platform was actually speedier than humans.

This time, Insilico seems to have avoided those pitfalls, says Cernak.

“I approached it with caution, because I think there’s a lot of hype in AI in drug discovery. I think Insilico’s been involved in hyping that, but I think they built something really robust here,” Cernak says. “I’ll be excited for a new world where, from target discovery to clinical trials, 2 years is the new timeline to achieve that.”

And as former DeepMind engineer and current PhD candidate at the Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory Jason Yim notes, there’s the possibility for Insilico and its contemporaries to build on this work: “The exciting prospect is how the AI pipeline can learn from its previous drug discovery cycles to potentially make this even faster,” Yim says.

Publishing in a peer-reviewed journal is also good business for Insilico. The company develops its own drugs, but it often licenses them to partners like Exelixis, which paid Insilico $80 million up front for an AI-generated cancer drug last year.

Publishing on a drug that’s still experimental comes with risks, but the main benefit of transparency outweighs those risks, Zhavoronkov says: it brings in potential customers.

It also makes a business case for small molecules, which some pharma companies are scaling back on in light of the US’s Inflation Reduction Act. Use AI to speed up the process, and you can still make returns before drug pricing kicks in, Zhavoronkov says.

“We are basically a factory of drugs,” Zhavoronkov says. “We don’t want to develop and commercialize them all ourselves. We want to keep being a factory.”



This article has been sent to the following recipient:

Chemistry matters. Join us to get the news you need.