Fig. 7: Benchmarking DOLPHIN for improved sensitivity and accuracy in alternative splicing detection.
From: DOLPHIN advances single-cell transcriptomics beyond gene level by leveraging exon and junction reads
The performance of DOLPHIN in detecting alternative splicing (AS) events was evaluated using real and simulated datasets. a–e Results from single-cell RNA-seq datasets, comparing DOLPHIN to SCANPY, scVI, Outrigger, MARVEL, BRIE2, SCASL, scASfind, and scQuint. DOLPHIN consistently outperformed these methods by leveraging exon-level information. a DOLPHIN identifies more AS events, including exon skipping (ES) and mutually exclusive exons (MXE), than all other methods. b Higher Pearson correlations between single-cell and pseudo-bulk datasets indicate DOLPHIN's improved splicing quantification, based on N = 80 bootstrapping replicates (10 per cell type across 8 cell types, using different random seeds; technical replicates). c Higher Adjusted Rand Index (ARI) scores for cell clustering based on Percent Spliced-In (PSI) values reflect DOLPHIN's ability to capture biologically meaningful splicing patterns, based on N = 100 bootstrapping replicates using different random seeds (technical replicates). d DOLPHIN detects more differentially spliced events per cell type based on Outrigger analysis, demonstrating its sensitivity to cell-type-specific splicing variations. e Higher ARI and NMI scores for splicing modality profiles, derived from Anchor analysis, further confirm DOLPHIN's ability to resolve splicing-driven cellular heterogeneity, based on N = 100 bootstrapping replicates using different random seeds (technical replicates). Boxes indicate the interquartile range (IQR, 25th to 75th percentile), with the line inside each box representing the median. Whiskers extend to the most extreme data points within 1.5 times the IQR from the quartiles. P values from one-sided Student’s t-tests: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; n.s. not significant. Exact P values are provided in the source data. f, g DOLPHIN using simulated datasets with known ground truth. f Receiver Operating Characteristic (ROC) curves show higher area under the curve (AUC) values for DOLPHIN, confirming its superior accuracy. g Precision-Recall (PR) curves demonstrate DOLPHIN's higher area under the precision-recall curve (AUPRC), indicating better precision and recall for splicing event detection. Source data are provided as a Source Data file.
