Abberant junctions by tophat

[RockChalkJayhawk]

I have tried to align my paired end RNA-Seq reads to the genome using Tophat. I ran a sample dataset from the SRA (SR018268_1 and _2) and the data looked fine. However, when I run my datasets, I get a lot of spurrious junctions. In the attached example, I show the junctions and coverage for one sample. All the exons map beautifully and have coverage > 200X, but the junctions between exons were not determined for almost all of these exons are not joined and the majority of "junctions" (>80%) in the dataset are intergenic (or intragenic) even with low coverage. For exampl, the far left junction is supported by 92 reads, the middle by 83, and the right by 2.

I have tried to manipulate the alignment parameters such as -r set to either 165 or 41. These correspond to 230 bp DNA identified from the bioanalyzer minus the inner distance alone (230-35-35=165) or including the primer sequences (230-35-35-119=41). This didn't really change things much.

So my questions are:
1) Why aren't these junctions being called by tophat?
2) Why would the junction on the right show up?
3) How do I get past this?

[Cole Trapnell]

How long are these reads?

[RockChalkJayhawk]

these are 2 x 35 bp reads. I also don't know if this matters, but my mapping qualities from the SAM files are:
632465 0
368741 1
38907221 255
1170126 3


Does this matter?

[RockChalkJayhawk]

Also, when I was trying to figure all this out, I made a fastq file that was only 100K long to troubleshoot and I ran into another problem. There are instances where junctions (true ones) appear only when the small dataset is used and not when the full dataset is used. Otherwise, they are exactly the same.

For instance, this figure shows no junctions when this gene is sequenced 33,333 times, but by subselecting and mapping with only 45x coverage, most of the exons are joined together.

Where did they go in the full analysis?

I am using the following code:

tophat -r 41 -p 6 --solexa1.3-quals hg19 sequence_1 sequence_2