Tumour cells survive inside a narrow band of mitotic instability — enough to keep evolving, not so much that division fails outright. Healthy cells sit nowhere near that edge. We are building a live-cell foundation model to find the perturbations that exploit the gap.
The instability that makes a tumour dangerous is also what makes it fragile. A perturbation that adds a little mitotic stress can tip one population and not the other.
Taxanes and vinca alkaloids are spindle poisons — they work by breaking mitosis. The trouble is that they break it in every dividing cell, so the dose is limited by what the patient can survive rather than by what the tumour needs. The suffering is not a side effect of one bad drug. It is what happens when a therapy cannot distinguish a tumour cell from a healthy one, and when finding one that can takes a decade.
1 in 5people develop cancer in their lifetime, with 20.6M new cases in 2024 rising to 34M by 2050.
3.4%of oncology programmes entering Phase 1 reach approval, after 10–15 years and upwards of $2B each.
~90%of solid tumours are aneuploid — already dividing under stress that healthy tissue never experiences.
Every existing method destroys the cell it measures. Sequencing lyses it; high-content imaging fixes and stains it. That means you can compare populations, but never a cell to itself — and the causal link between a perturbation and a fate is lost. Live-cell video keeps the cell alive, so we can follow one mitosis through to its own daughters.
Live-cell video of mitosis — brightfield, H2B and tubulin, imaged every ten minutes. Lineage tracing links each baseline division to its own perturbed daughters.
A mitotic perturbation response atlas across four cell contexts, where every response is paired to its own control rather than to a separate population.
A foundation model trained on that pairing, learning the direction that moves a cell from a failing division toward a healthy one.
Extrapolation across tens of thousands of genes and molecules, ranked by the gap between tumour and healthy response. Only the top of the list reaches the bench.
Each alternative gives up at least one of the four properties you need in order to rank a perturbation by selectivity.
| Same cell, before & after |
Mitotic dynamics |
Cell survives measurement |
Scales beyond what’s measured |
|
|---|---|---|---|---|
| Brute-force wet-lab screening | ||||
| Single-cell RNA-seq models | ||||
| Fixed-cell high-content imaging | ||||
| Live perturbation lineages |
Sinai Health in Toronto — named Canada’s first Nikon Center of Excellence for cellular imaging in 2021.
Selected for the 2026–27 cohort, working with mentors through CDL’s objective-setting programme.
Age-aware control cells extend the same platform into an atlas of rejuvenating perturbations and age-specific vulnerabilities.
We are early, and the useful conversations right now are with people who know this problem from the inside.