Fig. 4From: Implicit-descriptor ligand-based virtual screening by means of collaborative filteringAVE bias. The image to the left visualizes the comparison of AVE Bias across 4 sets of training and validation data. The figure shows that the 4 training and validation sets are randomly split with similar bias measures across the sets, thereby minimizing the impact of variance across our study. The image to the right shows the spread of the AVE Bias for each target in the dataset. The distribution of the AVE Bias from − 0.5 to 1.5 (where closer to 0 indicates no bias) across targets facilitates the study of the resilience of the collaborative filtering algorithm to the impact of such a biasBack to article page