Disparate numbers in IGV sashimi plot and junctions.bed

[krespim]

I trying to detect widespread splicing defects and amongst other things I ran tophat to find new splice junctions on my libraries (PE, 100 bases):

tophat -p 12 \
-r 200 --library-type fr-firststrand  \
--transcriptome-index=$bowtie_tindex \
-o $outputdir \
$bowtie_gindex \
$fastq_R1 $fastq_R2

Afterwards, the junction.bed files were merged and used as input for DESeq2 (condition vs control, 2 replicates). Then I tried to look at some of the significant events in IGV to get some nice sushi plots... and here lies my problem. This event, and there are more, has a clear pattern, being expressed in one condition and not in the other:

chr17_24721664_24721852_chr17_24721714_24721802 0 0 264 591

However, when I look at that junction in igv the two libraries that have no expression have a high count:



Is this something to do with the way IGV counts reads in that junction, am I missing something, or is it all wrong?