BRIE: transcriptome-wide splicing quantification in single cells

Abstract

Single-cell RNA-seq (scRNA-seq) provides a comprehensive measurement of stochasticity in transcription, but the limitations of the technology have prevented its application to dissect variability in RNA processing events such as splicing. Here, we present BRIE (Bayesian regression for isoform estimation), a Bayesian hierarchical model that resolves these problems by learning an informative prior distribution from sequence features. We show that BRIE yields reproducible estimates of exon inclusion ratios in single cells and provides an effective tool for differential isoform quantification between scRNA-seq data sets. BRIE, therefore, expands the scope of scRNA-seq experiments to probe the stochasticity of RNA processing.

Keywords: Differential splicing; Isoform estimate; Single-cell RNA-seq.

Fig. 1

A cartoon of the BRIE method for isoform estimation. BRIE combines a likelihood computed from RNA-seq data (bottom part) and an informative prior distribution learned from 735 sequence-derived features (top)

Fig. 2

BRIE improves isoform estimates by using an informative prior on simulated data. a, b At very low coverage RPK=25, a scatter plot between the estimates of the exon inclusion ratios by BRIE and the simulation truth. a BRIE.Null uses five random uniformly distributed features to learn the prior. b BRIE uses one correlated feature with Pearson’s R=0.8 to the truth to learn an informative prior. c Pearson’s R between truth and estimate by BRIE, BRIE.Null, and three other methods for different coverages. RPK reads per kilobase

Fig. 3

BRIE improves splicing estimates by using sequence features. a–c Pearson’s correlation between bulk and single cells on exon inclusion ratio ψ in HCT116 cells. Scatter plot of ψ estimates by DICEseq (a) or estimated by BRIE (b). Box plot for all methods (c) in 96 cells. d–f Pearson’s correlation between single-cell pairs. Scatter plot of ψ estimates by DICEseq (d) or estimated by BRIE (e). Box plot for all methods (f) in 4608 cell pairs

Fig. 4

Detection of differential splicing between cells. a Percentage of differential splicing events between human HCT116 cells, detected by MISO, rMATS, BRIE, and its mode with shared weights (i.e., BRIE.share) with different thresholds. MISO and BRIE use the Bayes factor (BF) and rMATS uses false discovery rate (q value). b Percentage of differential splicing events between mouse early embryonic cells at day 6.5 or day 7.75. The threshold is BF>10 for MISO and BRIE, and q<0.05 for rMATS. The diamond indicates pooled reads of 20 cells in each group. c An example exon-skipping event in DNMT3B in three mouse cells at 6.5 days and three cells at 7.75 days. The left panel is a sashimi plot of the read density and the number of junction reads. The right panel shows the prior distribution as a blue curve and a histogram of the posterior distribution in black, both learned by BRIE. For the histogram, the red line is the mean and the dashed lines are the 95 % confidence interval. BF Bayes factor