A century ago, Theodor Boveri proposed that cancer begins when cells acquire abnormal chromosomes. It was an elegant hypothesis, but testing it at scale was impossible with the technology of his time. Now researchers at EMBL Heidelberg have built an AI system that can hunt for these chromosomal errors across tens of thousands of cells per day, finally quantifying how often healthy cells go wrong.
The answer: roughly one in ten cell divisions produces chromosomal abnormalities. When a key tumor suppressor gene is broken, that rate nearly doubles.
What MAGIC Does
The system is called MAGIC — Machine Learning-Assisted Genomics and Imaging Convergence. It works like an automated game of microscopic laser tag.
First, a machine learning algorithm scans live cell samples under the microscope, hunting for micronuclei: small compartments that form when bits of DNA get separated from the main nucleus during cell division. Micronuclei are the visible signature of chromosomal chaos, and cells that contain them are more likely to develop additional abnormalities.
When the algorithm spots a cell with micronuclei, it triggers a laser that bathes that cell in light. The light activates a photoconvertible dye that permanently marks the cell. Scientists can then isolate these tagged cells for detailed genomic analysis using single-cell sequencing.
The beauty of the approach is its speed. MAGIC can analyze close to 100,000 cells in less than a day. Previous methods required scientists to manually search for micronuclei under the microscope — tedious work that limited how many cells could realistically be examined.
The Numbers
Armed with this throughput, the EMBL team — led by Jan Korbel with Marco Cosenza as the primary developer — could finally measure something that had been impossible to quantify at scale: how often do normal cells produce chromosomal abnormalities?
The answer was higher than expected. Slightly more than 10% of cell divisions in healthy tissue spontaneously generate chromosomal errors. Most of these are likely cleared by the body’s quality control mechanisms. But they establish a baseline for how often things can go wrong even when everything seems normal.
When the researchers examined cells with mutations in p53 — the tumor suppressor gene that’s broken in roughly half of all human cancers — the error rate nearly doubled. P53 normally helps cells detect and correct chromosomal problems. Without it, errors accumulate.
Why This Matters
The findings validate Boveri’s century-old hypothesis with modern molecular evidence. Chromosomal instability isn’t just associated with cancer; it appears to be woven into the fabric of how cancer starts.
But the implications go beyond academic vindication. If one in ten divisions produces chromosomal abnormalities even in healthy tissue, understanding what determines whether those errors get corrected or become dangerous could be key to prevention. The cells that escape quality control and go on to seed tumors might have specific signatures that MAGIC could help identify.
The research also demonstrates what happens when AI removes a manual bottleneck from science. The questions Boveri asked in 1914 were good ones. The answers just required examining more cells than any human could count.
The Fine Print
MAGIC currently focuses on micronuclei as the visible marker of chromosomal problems. There may be other types of errors that don’t produce these structures and would go undetected. The system also requires specialized microscopy equipment with laser capabilities, limiting where the research can be replicated.
The 10% error rate refers to chromosomal abnormalities broadly, not to cancer-initiating events specifically. Most cells with errors either die or are eliminated by the immune system. The chain of events from chromosomal chaos to actual cancer involves many additional steps that this research doesn’t address.
Still, having actual numbers where before there were only estimates changes what questions can be asked next. A century after Boveri, the theory is no longer just theory.