Table 2.

Most fit generative models learned by HIBACHI, trained on the propensity score matched UKBB genotype dataset. Along with the model, HIBACHI produces a synthetic version of the training dataset constructed using that model. Higher fitness scores indicate better approximation of the training dataset. Arithmetic operations are applied to the values (0, 1, or 2) comprising the input dataset – for example, “X₁₀!” indicates “the factorial of the value representing SNP X₁₀”. “XOR” and “AND” are logical Boolean operations, and ‘mod’ is the modulo operation.