Simulation Analysis
In order to benchmark the link quantifications generated by Altrans, we conducted a simulation analysis using the Flux Simulator software.17 We simulated an RNA-sequencing experiment with 50 million reads with the GENCODE v.12 annotation18 reflecting cases where we have a perfect annotation describing all the observed transcripts in the data. Additionally, we introduced novel transcripts, made up of existing exons of a gene, into the annotation. This was achieved by creating novel combinations of exons of a gene while checking for compatibility of the randomly selected exons (non-overlapping, order matches the genomic order, and where a UTR or first or last exon is not an internal exon) and keeping the distribution of number of exons of the random transcripts similar to that of the known transcripts. We then simulated 5 cases with 50 million reads where the novel transcripts accounted for 5%, 10%, 25%, 50%, and 75% of all transcripts, reflecting cases where the annotation is not perfect. Altrans, Cufflinks,11 and MISO13 were run on these 6 simulated datasets using the standard GENCODE v.12 annotation. In each simulation, the “correct” quantification of a transcript is taken as the RNA molecule count that the Flux Capacitor used to simulate reads for a given transcript. We have converted these “correct” transcript quantifications and the measured transcript quantifications of Cufflinks and MISO to exon link quantifications, as described in the previous section, and correlated the simulated expected link quantifications with the measured link quantifications for the three programs in the six simulation scenarios, using links where there were overlapping reads or links that were quantified in both the simulation and the given program. We measured the concordance between the simulated and measured quantifications via Spearman’s correlation. The estimates of novel splicing in a dataset are done through counting the number of uniquely mapping split reads. We then take junctions that are represented by at least eight split reads and check whether this junction is present in the annotation.