Figure 4.

Hyperparameter sensitivity analysis ERBB network. The Figure shows sensitivity of the AUC _ROC (panel A) and AUC _PR (panel B) of the ERBB network (with literature prior) with respect to changes in model hyperparameters. Model parameters μ, s, q, γ, r ₀ and s ₀ where changed by up to ±50%, and inference was repeated for each parameter choice. Shown are resulting values for the area under the receiver operator characteristic (AUC _ROC) and the area under the precision-recall (AUC _PR) curves. The analysis indicates that inference results are reasonably robust to changes in model hyperparameters, with the stochasticity parameter γ being the most critical parameter for inference performance.