Explore Kallisto-esq skipping & Equivalence Class pre-computation

[rob-p]

Right now, we're using Geet's "dirty-mapping" strategy to try and determine the origin of reads (dirty mapping treats the left and right-most hashable k-mer of each read as "mini" paired-ends, and uses the information about the k-mers to which they map to figure out where the read goes). It would be useful to explore 2 particular ideas used in the psudeo-alignment procedure of Kallisto.

• Break the transcriptome into simple paths of the de Bruijn graph. This will give us a data-driven rule to compute exactly how far one might need to "skip" along a read to find another informative k-mer. Note that while this idea is conceptually simple, there are definitely some tricky issues (e.g. we should properly handle the case where a transcript begins / ends in the middle of a simple path, as would happen with e.g. alternative start sites etc.). This means that a simple path in the dBG may contain multiple transcript cover-classes depending on where transcripts begin / end.
• Pre-compute unique labels for every combination of transcripts we see in the dBG cover. Note, these are not exactly the same as the equivalence classes that e.g. Salmon uses during quantification since, there, intersecting the hits for the ends of a paired-end read could lead to unique combinations of transcripts. However, here, the idea would be to essentially label all of the transcript cover-classes with unique identifiers. This would make the intersection of the transcript sets for the "simple" case (both reads come from the same set of transcripts) trivial --- just comparing the equality of the labels.

[rob-p]

The pre-computation of equivalence classes is done, but we should still explore similar "skipping" rules --- for now we use trivial skipping rules.

[rob-p]

Ok, a more sophisticated skipping heuristic has been implemented (based on informative skips computed during index creation). There may still be significant room to improve the accuracy and efficiency of this heuristic, but that warrants a new issue; I'm closing this one.