The Loraine Lab has developed alternative splicing RNA-Seq analysis tools called "find_junctions" and "arabitag" that we've used in conjunction with statistical testing to detect when splicing changes in an experiment.

This software works OK but interpreting results when one gene has multiple alternative splicing choices is difficult. Another potential weakness is the statistical test we use to detect a change across conditions. To test for a change, we compare proportion means using a t-test, after converting the proportions data to something more normal, using a transformation. In practice, this works OK, but things get confusing when we have multiple splicing patterns affect the same region of a gene, as we end up comparing each option pairwise against each other option. This is not a deal-breaker, but it would be nice if we could perform a single test even for alternative splicing choices where there are more than two options. There may be better software or better methods we can use instead.

However, one thing to always keep in mind is that we mainly care about understanding how and if splicing is regulated, and in detecting changes in the splicing machinery's function. In this case, what matters ultimately the frequency with which individual splice sites get used relative to alternative options, and what causes this frequency to change. Currently, we assess splice site selection frequency by counting relevant reads and ignoring both irrelevant reads, such as reads that align outside the differentially spliced regions, or reads we can't assign to a single choice. Moreover, because of the complexity of splicing and because of the unknowableness of reads mapping far away from a given splice site, we ought to focus our attention on the reads for which we are most confident in their mapping. What we care about the most is whether or not the underlying frequency of splice site usage changes between treatments or conditions. It's nice if we can know what that frequency is most likely to be, but a very rough estimate of the actual value is fine. Again, we care about the fact of a change, because the change signals that something biologically interesting might be happening.

For this task, let's scan the papers found in the literature review results (see: IGBF-3041) to identify alternative splicing analysis software packages and determine: Which of these, if any, should we explore as an alternative to the find junctions pipeline?

Also, let's investigate the statistics literature to determine if the test we need is already developed for an analogous setting that we can deploy for ourselves. We already have software that counts and reports relevant read alignments (see: arabitag), and so really we only need a good way to test for differences in splicing across conditions.