Fig. 3.

Effect of number of active compounds (Class size) and intra-target Tanimoto similarity (Class Tc) on F₁ Score performance. Sphere excluded targets perform better in comparison to under-sampled targets. Targets with high Tc and consisting of more than 40 active compounds perform well overall. In contrast, targets comprising a small number of active compounds, e.g. 40 or fewer, display a large variation in target similarity and performance. Marker colour intensity represents depth in 3D space (transparency indicates markers that are further away). Class Tc is calculated using the average of the top 5 nearest neighbours of the training compounds