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

GPC generates artificial genomes with 20% higher imputation accuracy and better privacy preservation than existing models, capturing complex genetic dependencies.

• Why it matters: Accurate and privacy-preserving generative models are crucial for population genetics research, especially when data sharing is restricted and long-range genetic dependencies are complex.
• What they did: The authors developed Genetic Probabilistic Circuits (GPC), a deep generative model based on hidden Chow-Liu trees, capable of modeling long-range SNP dependencies and supporting exact probability computations.
• The result: GPC produces realistic artificial genomes that improve imputation, especially for rare variants and underrepresented populations, while maintaining data privacy, enabling safer and more accurate genetic analyses.