Realistic artificial genomes that capture complex patterns in human DNA

What is it about?

Human genetic data are essential for studying disease, but they are also personal, and privacy rules often stop researchers from sharing them. One way around this is to create artificial genomes: synthetic data that reproduce the patterns of a real population without belonging to any real person. We built a model, called GPC, that learns which positions in the genome are most strongly linked to each other and connects them directly, even when they are far apart, instead of assuming that only neighboring positions matter. Because the model's probabilities can be calculated exactly, it can also fill in a person's missing genetic information directly, without first creating artificial genomes.

Featured Image

Photo by MJH SHIKDER on Unsplash

Photo by MJH SHIKDER on Unsplash

Why is it important?

Artificial genomes are only useful if they are realistic and do not leak information about the real people used to make them, and existing methods struggle to do both. Across data from the 1000 Genomes Project and the UK Biobank, GPC produced artificial genomes that reproduce the structure of real populations, filled in missing genetic variants more accurately than other machine learning approaches, and revealed less about the individuals it was trained on. The gains in accuracy were largest for rare variants and for populations that are poorly represented in public reference collections, which are exactly the cases where better tools are most needed.

Perspectives