cgsplice is a set of tools to support direct measurement of splice junctions in RNA-seq data. cgsplice works by finding and counting reads that span splice junctions. It does this by searching for reads with an "N" gap in their CIGAR alignment string. Next, for all junctions, it determines the donor and acceptor sites and tallies how many reads span each donor/acceptor pair. Finally, using these counts for multiple samples, it will calculate a permuted p-value for how likely it is that a given donor or acceptor had differential splicing between two conditions. Reads supporting a cassette exon, but that don't cross a junction are not included in this model.
junction-count command will count how many reads span any particular splice junction. Specifically, each junction is split into donor/acceptor pairs. Then for all junctions with a common donor or acceptor, the number of supporting reads is calculated. Only reads that have split-mapping (N gaps in the CIGAR string) are used.
Counting is done in a gene model-free method, allowing for discovery of unannotated transcripts (if a suitable aligner is used). In addition to counting reads that span junctions, reads that represent retained introns can also be counted.
splice-diff command takes a series of
junction-count outputs and performs a permutation analysis to determine which splicing donor/acceptor sites exhibit differential splicing between two conditions. The statistical metric used is a Student's t-score with Binomial variance. Samples are permuted across different conditions to calculate a p-value for each donor/acceptor.
combine-events command merges donor/acceptors into complete splicing events, such as cassette exons. Only donor/acceptor that meet certain criteria (FDR, effect-size) are merged.
bam-stats command can be used to determine how many junction-spanning reads are present in any given sample, among other things. These QC metrics can be used to assess how useful a given BAM file is for junction analysis.
cgsplice is implemented in Java and distributed as a self-executing fast-JAR file. This requires only a recent version of Java to be installed on the system. Otherwise, cgsplice acts as a normal command-line program.