Creating artificial human genomes using generative neural networks

Generative models have shown breakthroughs in a wide spectrum of domains due to recent advancements machine learning algorithms and increased computational power. Despite these impressive achievements, the ability generative create realistic synthetic data is still under-exploited genetics absent from population genetics. Yet known limitation field reduced access many genetic databases concerns about violations individual privacy, although they would provide rich resource for mining integration towards advancing studies. In this study, we demonstrated that deep adversarial networks (GANs) restricted Boltzmann machines (RBMs) can be trained learn complex distributions real genomic datasets generate novel high-quality artificial genomes (AGs) with none little privacy loss. We show our generated AGs replicate characteristics source dataset such as allele frequencies, linkage disequilibrium, pairwise haplotype distances structure. Moreover, also inherit features signals selection. To illustrate promising outcomes method, showed imputation quality low frequency alleles improved by augmentation reference panels RBM latent space provides relevant encoding data, hence allowing further exploration solving supervised tasks. potential become valuable assets studies providing yet compact representation existing high-quality, easy-access anonymous alternatives private databases.