From 300 Genes to 3 Targets: An LLM-BioGPU Case Study in ALS
An R&D team hands an LLM a list of 300 ALS-associated genes and asks a hard question:
Which biology is actually driving disease - and which targets are worth pursuing?
The LLM doesn’t answer with another list.
It calls the Biology GPU - a runtime that visually computes inside cells - and enters a closed-loop of experimental reasoning that moves from association to mechanism, and ultimately, to novel targets.
The Biology GPU is the biology runtime for AI enabling models to visually compute inside cells and create a closed experimental reasoning loop between hypothesis generation and functional biology at AI scale.
Built on our proprietary technology that directly visualizes biological processes and pathways, the Biology GPU runs a Visual Biology Model trained on more than 2 billion proprietary cellular pathway images.
In practice:
- AI models generate hypotheses from genomics, proteomics, or literature.
- The BioGPU converges these hypotheses into biological processes and runs them visually inside cells. Functional biology at AI scale.
- Together, this creates a closed loop; reasoning experimentally and visually on the biology, identifying which pathways are active, where compounds act, and providing a direct path to actionable targets and strategies for their modulation.
- Even starting from public data, the Biology GPU discovers novel biology. As an example, in ALS, an LLM connected through the BioGPU Agent to the BioGPU started from 300 disease-associated genes, converged experimentally on two biological processes that defined the disease signature, and then iterated this closed reasoning experimentally until it identified 3 novel targets.
Imagine what the Biology GPU could enable your AI models to discover when running not on literature-based hypotheses, but on the unique ones generated from your proprietary genomics and proteomics assets. You’d be uncovering causal biology and novel targets no one else could reach.