Fig. 2

ASP/L is positively correlated with organism complexity in the FirstSpeciesSet. (A) The 10 columns are combinations of four datasets (D1–D4) and three tools (LeafCutter, SpliceTrap, and StringTie). StringTie was not run for the D2 and D4 datasets because it failed to construct full-length transcripts as a result of the uneven distribution of the sampled reads. The 30 rows are combinations of the three numbers of genes used for ASP/L calculation (all genes, Bin 500, and Bin 100), two statistics for alternative splicing level [alternative splicing prevalence (ASP) and alternative splicing level (ASL); for the binned genes, we also calculated the predicted ASP (PASP) and ASL (PASL)], and three correlation calculation methods [Spearman’s correlation, linear regression, and phylogenetic general least squares (PGLS) regression]. The colors of the cells represent the Spearman’s correlation or adjusted R-squared value of the linear regression and PGLS regression. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, +P ≤ 0.1, and −P > 0.1. (B) Distribution of the P values in (A) across the four datasets and three tools. (C) Comparisons of cell type number (CTN) and data quantity in the D1 and D3 datasets of the 14 species. (D) Distribution of the correlation differences between the Spearman’s ρ between ASP/L and CTN, and between ASP/L and data quantity. (E) Distributions of P values calculated using the three gene selections (all, Bin 500, and Bin 100). (F) Distribution of P values calculated using the four measurements of ASP/L. (G) Distribution of P values calculated using the three correlation methods. (H) Boxplots of Spearman’s ρ between ASP/L and CTN using either all 14 species or 13 species without Pvi. The mean ρ values with and without Pvi were 0.61 and 0.59, respectively