Abstract
Cancer originates from alterations in the genome, and understanding how these changes lead to disease is crucial for achieving the goals of precision oncology. Connecting genomic alterations to health outcomes requires extensive computational analysis using accurate algorithms. Over the years, these algorithms have become increasingly sophisticated, but the lack of gold-standard datasets presents a fundamental challenge. Since genomic data is considered personal health information, only a limited number of deeply sequenced cancer genomes are available for distribution. As a result, tool benchmarking is often conducted on a small set of genomes with uncertain ground truths, which makes it difficult to measure the accuracy of analytic workflows.
To address this issue, we developed a novel generative AI tool called OncoGAN to generate synthetic cancer genomes based on training sets derived from large-scale genomic projects by employing generative adversarial networks and tabular variational autoencoders. Our results demonstrate that this approach accurately reproduces the number and frequency of mutations and their characteristics. Furthermore, it captures the genomic position of the mutations following the patterns specifically found for each tumor, and it enables us to replicate tumor-specific mutational signatures. To evaluate the fidelity of the simulations, we tested the synthetic genomes using DeepTumour, a software capable of identifying tumor types based on mutational patterns, and demonstrated a high level of concordance between the synthetic genome tumor type and DeepTumour’s prediction of the type.
This tool will allow the generation of a large realistic training and testing set of cancer genomes containing known genome alterations. This represents an advance for computational biologists, who will now have access to a publicly available set of realistic synthetic genomes with no privacy concerns, which can be used to develop new algorithms, improve the accuracy of existing tools, and benchmarking.
Competing Interest Statement
The authors have declared no competing interest.