Fig. 3

Detailed results on simulation performance benchmarks. a The performance of all methods in the Levine_32dim dataset. The four panels correspond to MSE in mean (smaller is better), MSE in covariance (smaller is better), KL Divergence (smaller is better), and relative efficiency in pMSE (larger is better). The relative efficiency is measured using Cytomulate’s pMSE: values smaller than 1 indicate worse performance than Cytomulate. b Orders of each method across all datasets and all metrics. Orders are based on values from the metrics, and a higher order indicates superior performance in the given metric. c The average performance of methods averaged across all datasets. Orange colors indicate superior performance, whereas blue colors represent worse performance. The interpretations are the same as a. d Orders of all methods averaged across all datasets. Higher bars represent a higher average order for a given metric and method. Colors represent different simulation methods