GPC: An expressive and tractable deep generative model for genetic variation data.

Genetic Probabilistic Circuits (GPC) accurately reproduce population structure and linkage disequilibrium patterns in genetic variation data, outperforming other models in imputation accuracy.

• Why it matters: Accurate generative models are essential for benchmarking, testing hypotheses, and constructing reference panels in population genetics, but current models struggle with dependency expression, tractability, and privacy.
• What they did: GPC is a deep generative model based on hidden Chow-Liu trees represented as probabilistic circuits, allowing arbitrary tree structures over latent variables and enabling exact probability computations for AG generation and genotype imputation.
• The result: GPC generates artificial genomes that reflect real population features, improves imputation especially for low-frequency variants and underrepresented populations, and better preserves training data privacy, supporting practical applications with limited data access.