Adrian Grzybowski, Chief Science Officer and Co-Founder of AnuBio, is helping build an AI-driven drug discovery platform designed to predict how living tissues respond to drugs, genetic edits and disease before those interventions reach the clinic.
AnuBio is focused on a longstanding challenge in pharmaceutical R&D: the difficulty of predicting how a promising molecule will behave inside a complex, multicellular biological system.
The company’s mission is to move early drug discovery away from trial and error and toward a more predictive model. Its core platform, Trailblazer, is a foundation model designed to simulate how interacting populations of cells respond to different perturbations.
“We want to make early drug discovery predictive, instead of being trial and error,” Grzybowski explained.
Addressing the Prediction Gap
According to Grzybowski, many drug discovery programs fail because the models used early in development do not accurately reflect what happens later in a living organism.
A molecule may perform well against a defined target or in an isolated laboratory system, only for unexpected biological effects to emerge further downstream. This creates both a scientific and economic problem, as weak early prediction can allow costly programs to progress before failure becomes apparent.
AnuBio was founded to help close that gap.
The company is starting in immunology, where interactions between cells are particularly important. Advances in single-cell biology have generated unprecedented amounts of high-resolution data, but Grzybowski believes many existing AI approaches still fail to represent biology as an interconnected system.
Modeling Tissues, Not Cells in Isolation
Trailblazer is built around the idea that a tissue should be modeled as a network of interacting cells rather than as a collection of independent cell types.
Grzybowski contrasts this with architectures derived from language models. Transformer-based models are highly effective for ordered sequences such as text, but biological tissues do not operate like sentences.
“A tissue is not a sentence,” he said. “It is a web of cells influencing each other with no fixed order.”
In immune biology, for example, the behavior of an individual T cell depends heavily on signals coming from surrounding cells. AnuBio therefore aims to predict how the effect of a drug, disease state or genetic modification propagates across multiple interacting cell populations.
This multicellular approach, the company believes, could provide a more realistic view of treatment response and help researchers make better decisions about which programs to advance.
Building the Data and Validation Engine
One of AnuBio’s biggest priorities is expanding the quantity and quality of data available to train its models.
The company is developing its own in-house data generation capabilities, particularly for multicellular perturbation datasets that remain relatively scarce. Controlling how those datasets are generated allows AnuBio to improve consistency, quality and biological coverage.
The second priority is tightening the connection between computational prediction and experimental validation.
“A prediction is only as good as reality at the bench,” Grzybowski said.
By shortening the cycle between AI-generated predictions and wet-lab validation, AnuBio aims to continuously improve prediction accuracy and create a more effective feedback loop between computation and experimental biology.
Looking ahead, the company’s growth will depend on expanding those proprietary datasets, strengthening its experimental validation capabilities and demonstrating that Trailblazer can help drug developers identify promising programs—and terminate weak ones—much earlier.
If successful, AnuBio believes predictive multicellular modeling could improve not only the science of drug discovery, but also its economics by reducing the cost of failure and helping the industry make better decisions earlier in development.
