“…To further compare the ranking performance, we used an adaptation of the rank-biased precision (RBP) algorithm (see Supporting Information), commonly used to assess the ranking precision of ranking algorithms in web search engines. Our adapted RBP version yields a normalized score, from 0 to 1, giving a score of 1 when the ranking is perfect and 0 where the ranking is the worst possible, by weighting the importance of the hit order; i.e., the score drops more significantly if the order changes from [1,2,3,4] to [2,1,3,4] compared to from [1,2,3,4] to [1,2,4,3]. RBP-based ranking scores across prediction methods are shown in Figure 4 b, where statistically significant differences are observed between CCSBase, AllCCS, DeepCCS, and LinECFP compared to CCSP2.0 (Wilcoxon test) and also compared to KerasECFP (p-value < 0.0001, significance not shown in the Figure 4).…”