Deep Origin and BioAge report finding molecules for a new drug target computationally in under three months, with confirmation in BioAge’s laboratory

Deep Origin and BioAge report finding molecules for a new drug target computationally in under three months, with confirmation in BioAge’s laboratory

On October 14, Deep Origin announced a collaboration with BioAge Labs, which develops drugs for cardiovascular and metabolic diseases by targeting the biology of aging. A drug target is a protein that a drug is intended to act on. Deep Origin uses computational tools to identify molecules for BioAge’s targets, and BioAge tests their activity in the laboratory.

One of BioAge’s data sources is the Norwegian HUNT biobank. In 2025, the company announced a project to measure thousands of proteins in each of more than 17 000 samples collected over decades and relate those measurements to participants’ diseases.

Michael Antonov, co-founder of the VR company Oculus and chief executive officer of Deep Origin, described BioAge’s targets:

“BioAge identifies targets using data from people followed for decades, and many of these targets are new, with little known about their chemistry. Our virtual screening was built for precisely this task.”

When few molecules are known to act on a target, there is little to guide the search, so virtual screening relies on the protein’s structure. It searches a catalog of made-to-order molecules and selects those predicted to fit into a protein pocket, a cavity on the protein’s surface. Neural networks such as AlphaFold 3 were trained on known protein-molecule pairs and perform less well on unfamiliar cases. According to Deep Origin’s August preprint, they position the molecule correctly in the pocket in 75–88% of the most familiar cases and no more than 25% of the least familiar cases. In Deep Origin’s DODock software, a neural network proposes molecular poses, which a physics-based formula with just 80 parameters then refines. The authors explain that the small number of parameters limits the formula’s ability to memorize training examples. In the least familiar cases, DODock correctly positions more than half of the molecules.

In the preprint, Deep Origin’s computational tools and chemists searched for molecules targeting CD73, an enzyme that helps tumors suppress the immune response. They selected candidates from approximately 80 billion molecules and tested 183. A molecule counted as a hit if it reduced enzyme activity by half at a concentration below 500 micromoles per liter. Of the 183 molecules tested, 56 met this threshold (30,6%), and their structures showed little similarity to known inhibitors of the enzyme. In a screening campaign using the AtomNet neural network, this enzyme was among the most difficult targets: one of 335 compounds was confirmed as a hit (0,3%). The preprint’s authors acknowledge that differences in catalog size, protein preparation, screening strategy, and assay format may partly explain the difference in results.

According to the companies, Deep Origin identified several structurally distinct hits against BioAge’s first target in under three months, and BioAge confirmed them in its laboratory. Before laboratory testing, Deep Origin filtered candidates using criteria set by BioAge’s chemists so that the hits would be suitable starting points for further development. BioAge is moving the target into hit-to-lead development, in which a weakly active compound is refined into a more potent and selective molecule. Subsequent stages involve selecting a drug candidate and conducting clinical trials. Deep Origin is eligible for milestone payments, and testing of new hits against the specified targets continues.

In September, Garegin Papoian, co-founder and chief scientific officer of Deep Origin, discussed computational drug discovery tools in a Drug Target Review article:

“Making the process cheaper and faster is only part of what is needed. Industry teams will adopt these tools as their primary strategy only when they can trust that computationally identified hits will translate into success in the laboratory.”