Fig. 4
From: Effect of missing data on multitask prediction methods
Comparison of removal models on HTSFP subsets. Median MCC values relative to the model with complete training data are shown. Results from models that removed individual activity labels are shown in blue, while results from model that removed whole compounds are shown in red. Results for DNN (left) and Macau (right), as well as results for HTSFP5 (top) and HTSFP10 (bottom), are shown independently