Insilico's AI networks generate custom lead compounds for fibrosis in less than 50 days

computer language
Over 21 days, Insilico and its partners used its artificial intelligence programs to conceptualize and generate 30,000 novel small molecules that may work against fibrosis. The most promising six were selected after 25 more days. (Getty Images)

In the gold rush to bring artificial intelligence to the healthcare and biopharma industries, AI has long been pitched as a way to accelerate the pace of drug development and discovery. Sometimes vaguely and sometimes not, many companies have claimed their code can help early research get done quicker, deeper and cheaper. 

Now, Insilico Medicine may have hit pay dirt, demonstrating in a paper published in Nature Biotechnology that its computer networks could potentially shave years off of traditional hit-to-lead timelines.

Over 21 days, the startup and its partners used its AI programs to conceptualize and generate 30,000 novel small molecules that may work against fibrosis. Within 25 more days, they had screened out and synthesized the six most promising compounds, tested them in vitro for selectivity and metabolic stability and had the lead candidate go on to show favorable activity in live mouse models.

Webinar

How ICON, Lotus, and Bioforum are Improving Study Efficiency with a Modern EDC

CROs are often at the forefront of adopting new technologies to make clinical trials more efficient. Hear how ICON, Lotus Clinical Research, and Bioforum are speeding database builds and automating reporting tasks for data management.

“This paper is a significant milestone in our journey towards AI-driven drug discovery,” Insilico founder and CEO Alex Zhavoronkov said in a statement, describing the initial skepticism facing AI-generated chemistry. “Now, this technology is going mainstream and we are happy to see the models developed a few years ago and producing molecules against simpler targets being validated experimentally in animals.”

RELATED: WuXi AppTec to test AI-generated compounds from Insilico Medicine

The compounds aim to inhibit discoidin domain receptor 1, or DDR1, a receptor tyrosine kinase found on epithelial cells and involved in fibrotic diseases and disorders.

To design and build them, Insilico developed the machine learning approach GENTRL, named for generative tensorial reinforcement learning, which prioritizes the feasibility of synthesizing a compound, its effectiveness against a biological target and how distinct it is from other patented molecules found in medical literature.

The algorithm’s two-step process maps and explores the structures of new compounds within the parameters of a chemical space and uses a machine learning model trained on DDR1 and common kinase inhibitors. Going forward, however, the newly generated compounds may still need to be optimized for different medicinal properties, Insilico said.

“By enabling the rapid discovery of novel molecules and by making GENTRL’s source code open source, we are ushering in new possibilities for the creation and discovery of new life-saving medicine for incurable diseases—and making such powerful technology more broadly accessible for the first time to the public,” Zhavoronkov said.

Suggested Articles

Sanofi will look to pull back from its three-year-old relationship with Verily and their virtual diabetes clinic, Onduo.

NASH leaders weigh in on the need for a drug for the disease and the challenges in getting it to patients.

AstraZeneca is linking up with DeepMatter, a big data firm focused on achieving reproducibility in chemistry, to help improve its compound synthesis.