Table 3. Summary of benchmark (RB106, RB144 and RB198) results in terms of AUC (mean ± Std).

Benchmark [r]	RNABindR 2.0	PSSM	Sequence-based control	aaRNA
RB106	0.81	0.776 ± 0.001 *	0.803 ± 0.001 *	0.8251 ± 0.0009
RB144	0.81	0.782 ± 0.001 *	0.801 ± 0.002 *	0.830 ± 0.001
RB198	0.80	0.7696 ± 0.0007 *	0.7974 ± 0.0007 *	0.8343 ± 0.0004
Benchmark [p]	RNABindR 2.0	PSSM	Sequence-based control	aaRNA
RB106	0.74	0.721 ± 0.119 **	0.735 ± 0.109 **	0.765 ± 0.116
RB144	0.74	0.723 ± 0.118 **	0.733 ± 0.111 **	0.778 ± 0.105
RB198	0.73	0.716 ± 0.114 **	0.738 ± 0.106 **	0.784 ± 0.103

The corresponding ROC plots and AUC distribution patterns are given in Figure 4 (residue-based evaluation) and Supplementary Figure S13 (protein-based evaluation), respectively. RNABindR 2.0 is the best-performing sequence-based method from various approaches evaluated in the work (16). Its reported performance is listed. Sequence-based control method represents three sequence features of the aaRNA, which are adapted from the work SRCPred (12). In Benchmark [r], AUCs were measured on a protein-residue basis, and reported AUCs are the average results of five repetitions of five-fold cross-validation. The average AUC of the aaRNA method is significantly greater than that of the PSSM or sequence-based control method using a t-test. In Benchmark [p], AUCs were individually calculated for each protein chain, and a paired Wilcoxon test was applied to check whether the distribution of the aaRNA AUC is shifted to the right relative to that of the PSSM and sequence-based control. The significance of differences between the alternative methods and aaRNA is indicated by * for P-values < 10^-5 and ** for P-values < 10^-10.