2023-08-16
Modeling alternative splicing in single cells
Presentation by Karin Isaev
Is splicing bimodal in cells? Cells with more reads often have better range. What influences the observed splicing output? Cell size, single cell vs single nuclei, transcription rate (?), etc.
~4500 captured RNAs, ~1500-2400 captured genes in PBMC by different sequencing tools.
Smart-seq2: improved coverage, high level of mappable reads, higher sensitivity disadvantages: no early multiplexing, transcript length bias, lower throughput.
Split-seq: Higher throughput, improved coverage, lower cost, disadvantages: reduced sensitivity, less comprehensive alternative splicing information than full-length transcript methods.
Smart-seq3xpress: scalable single-cell RNA sequencing from full trnscripts. Is it possible to get high sensitivity and high throughput?
How to quantify splicing?
Splice junctions can be used for quantifying splicing. In the mouse reference, most junctions are “common” and found in all transcript isoforms for a given gene. In human, many more partially and fully unique junctions compared to mouse.
Split reads quantify presence or absence of intron cluster.
Tabula Muris dataset: Smart-seq + 10x sequencing on different cells in mouse. In bulk sample, 85,439 SJ before clustering, 9,390 after clustering. In pseudo-bulk sample (n = 432 cells), 101,801 SJ before clustering, 14,351 after clustering. In single cell, 9,731 before clustering, 481 after clustering.
Splicing events: - Skipping exon - Mutually exclusive exons - Alternative 5’ splice site - Alternative 3’ splice site - Retained intron - First exon - Last exon
How do you know the splice event from the PSI value?
What is Leaflet?
Input: BAM file (pseudobulk if single cells) Model: Binomial Mixture Model, Inference with CAVI. Output: Optimal K (# cell states), Proportion of each cell state in the data, junction usage values for each cell state (PSI), cell state labels for each cell (Z).
Cell states inferred by Leaflet overlaps with FACS classification in the data.