Missing noncanonical splice sites in TopHat align.

[wokai001]

We are searching for noncanonical splice sites.

We have produced RNA-seq of a set of 60 human samples and produced aligs against human DNA (Ensembl) using TopHat2 (versions 2.0.4 to 2.0.10).

From these aligns, we collected splice sites cumulative from these files (resulting in about 10E6 = 1 million different sites). We annotated these sites using Ensembl (74) by searching for the best matching hit for the splice sites and reverse - complemented the sequences on the (-) strand (some of which obviously are assigned to the wrong strand...).

When looking at the first two intronic Nucleotides on the DONOR site we find the following distribution:

AT CT GC GT
15887 72358 44801 825650

We miss any noncanonical splice sites. Does TopHat filter them out? Or is there another explanation?

[Brian Bushnell]

Wolfgang,

Your question is not entirely clear; I don't know whether the distribution you state is from some empirical source, or the Tophat alignments, or from annotation, or what. My impression is that Tophat alignments contradict Tophat alignments, which is clearly wrong, so please clarify.

Regardless, I am suspicious of any alignment that prioritizes known junctions or gene annotations. BBMap is capable of mapping RNA-seq data to the human genome and finding novel junctions, with zero bias toward known or canonical junctions, because it ignores that information (which is either good or bad, depending on your objective). For human RNA-seq, I would suggest a command-line like this:

bbmap.sh ref=hg19.fa in=reads.fq out=mapped.sam maxindel=400000 -Xmx23g xstag=firststrand intronlen=10 ambig=random

[wokai001]

Hi Brian,
we made the TopHat alignments without given annotations, just DNA sequence. Therefore, I would like to know if a bias in TopHat alignments is acknowledged. We see a lot of junctions which are not on annotated positions (otherwise we could not have 1E6 sites). But we don't see non-canonical sites.

Thanks for your advice on the aligner. I will check this.

Regards
Wolfgang