DeBruine’s research builds and trains generative AI to
construct what he calls a patient’s “cell atlas” from a small
portion of genetic data, such as a blood draw or skin biopsy. With
the “cell atlas,” DeBruine’s team can use AI to generate what whole
organism cell atlases of rare disease might look like even if the
rare disease was only measured in a single tissue or sample.
“Our models will also be able to generate data across
species,” DeBruine said. “For example, we aim to generate a portrait
of what a rare disease in humans looks like, based on data collected
from fish or mouse models of that disease.”
DeBruine said his team has developed models with nearly
1 billion parameters and trained on more than 80 million cells, each
with expression information on more than 60,000 genes.
He added the grant will support his team of four to
five undergraduate students and graduate computer science students
to perform full-time research. The students will develop the AI’s
algorithms and train these models.
“We believe our new models will harness the true power
of generative AI for genomics to achieve a new level of
understanding about human disease,” DeBruine said. “We look forward
to partnering with the single-cell data insights community at the
Chan Zuckerberg Initiative to make our models accessible to every
genomics researcher.”
