Table 2.

Performance evaluation of Ψ-site predictors.

Mode	Method	Benchmarking data (AUC)				Average AUC
		Ψ-Seq	RBS-Seq	CeU-Seq	Pseudo-Seq
Full transcript	PIANO	0.957	0.978	0.914	0.972	0.955
	iRNA-PseU	0.679	0.727	0.721	0.708	0.713
	PPUS	0.700	0.721	0.724	0.705	0.713
	PseUI	0.631	0.710	0.610	0.585	0.634
Mature mRNA	PIANO	0.859	0.770	0.864	0.857	0.838
	iRNA-PseU	0.753	0.582	0.760	0.751	0.712
	PPUS	0.749	0.575	0.757	0.748	0.707
	PseUI	0.666	0.651	0.652	0.639	0.652

The table presents the performance of different Ψ site predictors achieved on independent human datasets with different technologies as a benchmark, and it is summarized from Supplementary Table S3 and S4 . Only the Ψ sites not previously used as training data were considered during performance evaluation, so the training sites and testing sites did not overlap. Because existing datasets overwhelmingly relied on polyA selection in RNA library preparation and intronic Ψ sites are likely to be underrepresented in the data, the performances were evaluated under two modes: full transcript and mature mRNA modes. In the mature mRNA mode, only positive and negative Ψ sites located on mature mRNA transcripts are considered, as previously described (Chen K,et al., 2019). Our new approach PIANO substantially outperformed competing approaches in accuracy.