How to build reference genome for targeted region(from a BED file) for alignment?

[serenaliao]

I am working with Sureselect methylseq data. Previously we used the whole reference genome for Bismark alignment with mapping efficiency 50%~60%.
Considering most of reads should be around targeted regions in this kit, it should be better to just map to targeted regions in the provided BED file.

Any idea how to build the reference genome base on a BED file with targeted regions? Thanks!

[aggp11]

You could use bedtools' FastaFromBed utility by providing it your BED file and the whole genome reference FASTA file.

[lh3]

Usually, it is recommended to map reads to the whole genome. These kits still have false positives, which will lead to mapping artifacts.

[kenietz]

I suppose it depends on the project. In some cases its better to construct a special version of the genome containing only the targeted regions.

In my case of amplicon seq, 50% of the reads are mapped all over the genome.

Well at least i will try to create a special version of the genome and then remap and compare the results.

[sxh]

Hi everyone,

I am interested in finding somatic mutations from whole genome sequencing of tumors. I am primarily interested in reads that land in known regulatory regions of the genome. As aligning the reads to the whole genome is taking 2-3 days, I thought that aligning to just the regions of interest would cut down this processing time by a lot.
However, I'm also concerned about false positive alignments like @lh3 mentioned, that may result if the actual alignment is outside of my custom reference genome, or if the actual alignment is to repeat regions that appear unique in my custom reference genome. I was wondering if anyone has tried this strategy before and if so, what parameters I should consider changing in the aligner to cut down risk of false positives?

Thank you in advance for your time!

[Brian Bushnell]

His advice still holds. Particularly when you are looking at things where you expect a large difference from the reference (tumors, bisulfite, etc) it's bad practice to map only to the part you're interested in. There is no way to avoid the false positives that incurs, so your results will be untrustworthy.

If alignment is too slow, you can use a faster aligner, or tweak the settings of the existing aligner, or use a computer with more cores, or split up the data and map on multiple computers. How are you doing the alignment?

[sxh]

Thank you Brian for the fast reply. I will look more closely at the parameters for different aligners to see how to speed things up. So far I have tried BWA and Bowtie2, with default parameters. I have also tried dividing the original .bam into 24 parts and doing the alignment on separate cores, but it is still taking more than 2 days.