Discussion Here we present a method, Altrans, for relative quantification of splicing events (Figure 1) to be used in population genetics studies in discovery of asQTLs. Because the phenotype is splicing ratios of exon links calculated from mapping of RNA-sequencing reads without modeling of transcript structure, it is a more direct estimation of splicing. We have assessed the performance of the Altrans algorithm versus the Cufflinks method both on simulated and biological data. The simulation analysis indicates that when the annotation perfectly describes the underlying isoform variety, Cufflinks performs better than Altrans. Because there is no easy way to generate junction annotations that is used by MISO, and because we needed to have a common annotation in all analysis (we could not use the junctions provided in the MISO website), we chose to quantify transcripts rather than junctions. Although MISO might perform better if we had quantified junctions, the analysis performed is equivalent to the one with Cufflinks, and still MISO underperforms compared to Cufflinks. The reason Altrans is worse when compared to Cufflinks in the presence of a perfect annotation is that Cufflinks quantifies transcript rather than exon links, i.e., it uses the total length of the transcript in quantifications, whereas Altrans uses only the observed reads that are linking a pair of exons. When we convert the transcript quantifications of Cufflinks into link quantifications, this means that all the links in a transcript will “borrow” information from other links of the transcript, whereas in Altrans all the links will be independently measured from the observed reads overlapping the link. Moreover, when the perfect annotation is available using transcript quantifications, as in the case of Cufflinks, then Cufflinks is a more accurate approach. However, the simulations also show that when there are novel transcripts, i.e., isoforms that are not represented in the annotation, the accuracy of transcript quantifications decreases for Cufflinks and MISO whereas Altrans quantifications do not suffer as much as the transcript quantifications. We estimate that in less well-studied transcriptomes like the human pancreatic beta cell transcriptome,25 the proportion of the links between exons that are novel would be high enough that using the known annotation can result in unreliable estimates. It is important to assess the performance of a method using biological data, and we applied Altrans and Cufflinks to the Geuvadis dataset7 with the specific aim of identifying asQTLs. We find 1,427 and 1,737 asQTL genes in the European population and 166 and 304 asQTLs in the Africans with Altrans and Cufflinks, respectively. Using two subsets from the European samples, we show that Altrans and Cufflinks achieve similar levels of replication. Altrans-specific asQTLs accounts for 45% of this method’s discovery, which we show is mainly due to it quantifying junctions that are not annotated in the reference. Moreover, these Altrans-specific asQTLs replicate as well as the common genes, indicating that they are probably true positives. The other reason for the method-specific asQTLs is the different types of alternative splicing events each method captures (Figure S4). Altrans is more powerful in capturing exon skipping events, whereas Cufflinks appears to be as powerful in capturing events in the ends of transcripts. This is an expected result given how each method works. Because Altrans is examining reads that link multiple exons, it will perform relatively poorly when a read pair has to extend over constitutive parts of exon groups if constitutive parts are larger than the insert size of the experiment, because there will be very few reads joining these types of exons. On the other hand, because Cufflinks uses all reads over a transcript, it will not fail to quantify these types of events accurately and this is reflected in the types of events each algorithm identifies. Furthermore, when we compare replication of results of one method by the other method and account for the overlap of observed links between the two methods, we find similar levels of overlap between the detectable discoveries for each method. The relevance of the asQTLs identified by both methods is confirmed by their significant overlap with functional annotations. This result, in the absence of a comprehensive list of asQTLs, shows that asQTLs that we are capturing reside in biochemically active regions of the genome, which reaffirms that we are capturing real biological signal. RNA sequencing allows us to comprehensively measure transcript diversity in different cells types at the population scale. However, quantifying alternative splicing from short read length RNA sequencing remains a challenge. This problem will be alleviated when technologies that would permit sequencing of full-length transcripts, like nanopore sequencing,26 become available, reliable, and are cost effective in population studies. Currently all methods have to infer quantifications of transcripts or splice junctions, and each method in doing so has its relative merits. Here we present a different approach to this problem, called Altrans, and show that it is sensitive and performs comparably to other methods. We show that it is capable of identifying thousands of asQTLs, many of which are missed by other methods. We believe it will prove useful in the search for alternative splicing QTLs in population genetics studies.