Insilico Medicine's AI Pairs AlphaFold Structures With Generative Chemistry to Find Novel SIK2 Inhibitors

Insilico Medicine, a clinical-stage drug discovery company built around generative artificial intelligence, has published a study in the July 13 edition of Bioorganic & Medicinal Chemistry describing a new set of inhibitors for salt-inducible kinase 2, known as SIK2. The company calls the work a significant breakthrough in combining several generative AI models with AlphaFold protein structures.

SIK2 is considered a potential target for both anti-inflammatory and anti-cancer therapy. It is highly overexpressed in 30% of human ovarian cancers, which is part of the reason researchers want a selective way to hit it.

The method paired two tools. AlphaFold, developed by Alphabet's DeepMind, has predicted protein structures across the entire human genome and is available as a free database, helping scientists model the structures of millions of previously uncharacterized proteins. Insilico's generative chemistry engine, Chemistry42, then proposes novel chemical structures tailored to those targets. In the SIK2 work, the team used AlphaFold-predicted structures to generate a series of hinge cores, the pieces of a molecule that anchor it to a kinase.

The path from there followed the usual sequence: molecular docking, synthesis, and biological evaluation. That produced a hit molecule against SIK2 with a novel scaffold. Further exploration yielded a compound that the authors say is more potent against SIK2 than previously reported inhibitors. It also showed excellent selectivity over other AMPK kinases, favorable in vitro ADMET profiles, and decent cellular activities.

Xiao Ding, PhD, Insilico's Senior Director and Head of Chemistry and a co-author, said the work shows that Chemistry42 and AlphaFold-predicted structures can identify a series of novel, potent and selective SIK2 inhibitors through structure-based design, and that it further demonstrates the strength of the company's Pharma.AI platform.

This is the second study in which Insilico has combined its generative platform with AlphaFold. The earlier one, in Chemical Science, was done with Alán Aspuru-Guzik, director of the University of Toronto's Acceleration Consortium, and Chemistry Nobel laureate Michael Levitt. It applied an AlphaFold2-predicted structure to Chemistry42 to generate an inhibitor for CDK20, a promising target in hepatocellular carcinoma. In that effort, 8,918 molecules were designed, and 54 with unique scaffolds and diverse hinge binder profiles were prioritized. A hit molecule was identified, and in a second round two compounds showed strong potency against the intended target.

Founder and CEO Alex Zhavoronkov, PhD, framed the research as evidence that AI systems can work together to produce new therapeutics where structural data is limited. He said the company is encouraged by the findings and their promise for finding potent new targets and molecules for diseases with high unmet need.

Insilico says it continues to add new technology to its platform, including large language models and quantum computing alongside AlphaFold. Its lead generative-AI-discovered drug, for idiopathic pulmonary fibrosis, recently advanced to Phase II clinical trials. The company reports two other clinical-stage programs and more than 30 drugs in its internal pipeline spanning cancer, fibrosis, immunity, central nervous system diseases, and aging-related diseases.

The SIK2 results are company-reported and preclinical. What to watch is whether the compounds are validated in vivo and whether they move toward clinical development.