Researchers at the Ontario Institute for Cancer Research (OICR) and the University of Toronto have developed OncoGAN, a new generative AI system described in Cell Genomics.

OncoGAN can simulate realistic tumour genomes across eight cancer types, including breast, prostate, and pancreatic cancers. By recreating authentic patterns of genetic alterations, these synthetic genomes can be used to benchmark genomic tests and enhance the algorithms that power precision oncology.

Studying tumour genomes and their DNA variations has already transformed our understanding of how cancer develops, driving the creation of advanced diagnostics and therapies. This work underpins precision oncology, where treatments are tailored to the unique biology of each patient’s tumour.

Current algorithms for analyzing cancer genomes face limitations because they are trained on a relatively small number of samples, with only a fraction publicly accessible. Many widely used tools were built on just a few dozen older genomes, leaving them unable to reflect the full biological diversity of cancer. Although newer sequencing data exists, access is often restricted due to patient privacy concerns.

“With OncoGAN, we are creating realistic genomes out of nothing—completely detached from real individuals, yet offering immense scientific value,” explains Lincoln Stein, acting scientific director at OICR and professor of molecular genetics at U of T’s Temerty Faculty of Medicine, and senior author of the study. “These synthetic genomes contain no personal health data, which means they can be freely shared without restrictions.”

Another key advantage of OncoGAN is that each synthetic genome comes with a known ‘ground truth’—a complete, error-free DNA sequence with all genomic variants clearly identified. In real genomes, establishing ground truth is nearly impossible due to biological complexity and the limitations of sequencing technology. As a result, existing analysis tools may be flawed because they were trained on incomplete or imperfect data.

By generating genomes entirely from scratch, OncoGAN provides researchers with fully verified DNA sequences that allow for more accurate genomic testing and analysis.

“Having access to genomes with a known ground truth means new algorithms can be benchmarked with absolute certainty about the correct answer,” says Ander Díaz-Navarro, postdoctoral fellow at OICR and first author of the study.

With better-trained tools to interpret tumour genomes, Stein notes, researchers will be able to unlock deeper insights that could reshape cancer care.

The more we understand the biological drivers of cancer, the better we can detect it early, treat it more effectively, and ultimately work toward preventing it altogether, Stein adds.

Source: https://temertymedicine.utoronto.ca/news/ai-generated-genomes-could-accelerate-precision-medicine-without-compromising-patient