PerTurbo - The Agentic Drug Discovery Copilot

PerTurbo is an agentic AI copilot for drug target discovery. Many of the hardest diseases to treat are driven not by a single gene, but by complex multicellular mechanisms, whole ecosystems of cells signalling to each other in tissue. PerTurbo reads that ecosystem from spatial transcriptomics and lets a researcher interrogate it in plain language. Ask "what should I target in this sample," and the agent ranks cell populations by their predicted effect on the disease, drills into the genes driving each, maps where that effect lands across the tissue, draws the underlying gene coupling network, and grounds every candidate against live records from Open Targets, DGIdb, Europe PMC, and ClinicalTrials.gov. Each answer plays out as a short, narrated sequence , with the scientist watching the reasoning, not a bare gene list. This targets the most expensive bottleneck in drug development. Bringing a single drug to market costs over a billion dollars and takes more than a decade, and the great majority of candidates fail, most because the biological target was wrong. Target identification is a multi-billion-dollar market growing at roughly 19% a year, yet it still leans on months of manual analysis. PerTurbo compresses that first, decisive step: it turns a spatial tissue sample into a ranked, mapped, evidence-grounded shortlist in minutes, de-risking what advances into costly wet-lab and clinical validation. Built on the Celcomen spatial model, with the agent running on AMD via a Fireworks LLM backend.