In silico generation of synthetic cancer genomes using generative AI

Area of Science:

• Computational biology
• Genomics
• Artificial intelligence

Background:

• Genomic alterations are crucial for understanding cancer and advancing precision oncology.
• Privacy concerns limit the sharing of deeply sequenced cancer genomes, hindering algorithm development and benchmarking.
• Accurate algorithms are essential for detecting genomic alterations, but their improvement is impeded by data scarcity.

Purpose of the Study:

• To develop a generative artificial intelligence (AI) model for creating realistic synthetic cancer genomes.
• To address the challenge of limited data sharing in cancer genomics due to privacy concerns.
• To generate privacy-preserving synthetic cancer genome datasets for benchmarking and improving analytical tools.

Main Methods:

• Developed OncoGAN, a generative AI model utilizing adversarial networks and variational autoencoders.
• Trained OncoGAN on large-scale genomic datasets to learn patterns of somatic mutations, copy number alterations, and structural variants.
• Validated the fidelity of synthetic genomes using the DeepTumour tool, assessing concordance with real tumor data.

Main Results:

• OncoGAN accurately reproduces key genomic alterations, including somatic mutations, copy number alterations, and structural variants.
• Generated synthetic genomes reflect cancer-specific mutational signatures and positional mutation patterns.
• Synthetic data augmentation improved the accuracy of the DeepTumour analysis tool.

Conclusions:

• OncoGAN effectively generates realistic, privacy-preserving synthetic cancer genomes.
• The synthetic data can be used to augment real datasets, enhancing the performance of cancer genome analysis tools.
• OncoGAN offers a solution for creating shareable datasets with known ground truths, facilitating benchmarking and advancing cancer research.