Fig. 3.

This figure displays the tuned max_G and max_X hyperparameters using the StARS method [22] across the 9 different simulation settings accounting for different numbers of average cells per subject and different levels of technical noise. The points represent the median tuned values and the bars represent the first and third quartiles across 50 iterations for each simulation setting. (A) displays the tuned max network size within $\mathcal{G}$, and (B) displays the tuned max network size within $\mathcal{X}$. Ideally, MOSCATO would tune max_G to 10 and max_X to 15 in order to select the proper network size according to Table 1