I have been trying to generate simulated reads from an interchromosomal breakpoint for some time now. I'm quite puzzled by what I am generating however. I've been reading various papers where a method was tested on simulated data, but I don't seem to be able to duplicate the simulations.
I wonder if anyone could see where I could be going wrong here:
1. I create a FASTA file by taking sequence from two separate chromosomes, (e.g. 8 and 15) and merging them at a known point. I make sure this FASTA file is limited in size to about 1000bp either side of the breakpoint.
2. Using the ART simulator (I've also tried wgsim) I generate reads with a known size, gap and stdev.
3. Align the simulated reads to the full reference (hg19) using BWA and look at the resulting alignments
For the full reference alignment (hg19) very few of the reads align to the chromosomes I generated the reads from. They will align to two different chromosomes (e.g. 1, X or 2, 17) which may not be surprising, but makes me question the read simulation. In fact, only 10% of the simulated reads align at all and these are all generated with high base quality scores. I'm not sure how important that is.
If instead I create a reference that is only from the chromosomes that I know these reads come from (e.g. 8,15), the reads still don't align discordantly. Instead, both reads in the pair will align in the same position on a single chromosome.
I'm at a loss here, there must be a good way to do this since so many published methods test on simulated data. Thoughts?
I wonder if anyone could see where I could be going wrong here:
1. I create a FASTA file by taking sequence from two separate chromosomes, (e.g. 8 and 15) and merging them at a known point. I make sure this FASTA file is limited in size to about 1000bp either side of the breakpoint.
2. Using the ART simulator (I've also tried wgsim) I generate reads with a known size, gap and stdev.
3. Align the simulated reads to the full reference (hg19) using BWA and look at the resulting alignments
For the full reference alignment (hg19) very few of the reads align to the chromosomes I generated the reads from. They will align to two different chromosomes (e.g. 1, X or 2, 17) which may not be surprising, but makes me question the read simulation. In fact, only 10% of the simulated reads align at all and these are all generated with high base quality scores. I'm not sure how important that is.
If instead I create a reference that is only from the chromosomes that I know these reads come from (e.g. 8,15), the reads still don't align discordantly. Instead, both reads in the pair will align in the same position on a single chromosome.
I'm at a loss here, there must be a good way to do this since so many published methods test on simulated data. Thoughts?