One of the advantages of paired end sequencing over single end is that it doubles the amount of data. Another supposed advantage is that it leads to more accurate reads because if say Read 1 (see picture below) maps to two different regions of the genome, Read 2 can be used to help determine which one of the two regions makes more sense. However, I don't understand how this second advantage pertains only to paired end sequencing. With single-end sequencing, couldn't you just use the read that comes right after read 1 to essentially get a longer net read?
Like you mentioned, having read pairs can help with alignment. Your FASTQ for a single-end sequencing run will look something like this:
@SEQ_ID_0001 AGCTAGCGCGGTTGGCTTAGCGACT + !''*((((***+))%%%++)(%%%% @SEQ_ID_0002 AGGTGGTTGTAGGGAAAAAAGTCTC + !''*((((***+))%%%++)(%%%% ...
There is not necessarily a relationship between
They could come from neigbouring regions in the genome or they could come from different species if that's what your experiment involves.
@SEQ_ID_0001 doesn't exactly give you information about where
@SEQ_ID_0001 could equally map to multiple locations in the genome with the same probability.
How do we decide where it "should" go?
We can randomly distribute multi-mapped reads or do some other fancy stuff, but that usually relies on facts about the data as a whole, and not something specific to the original DNA molecule that sequencing read came from.
If we have paired-end sequencing, then like your diagram shows, you know that read 1 (
R1) and read 2 (
R2; these have the same
@SEQ_ID) will relate to each other in some way.
@SEQ_ID_0001 R1 may be multi-mapped, but if
@SEQ_ID_0001 R2 isn't, then we know that
R1 should probably be relatively close to
Because of how you prepare your DNA samples before sequencing, you'll have some information about the insert length distribution.
You can use this information with sequence aligners to better identify where
R1 should be placed.
You can also do this if both
R2 are multi-mapped; certain pairwise alignments will be more likely than others, even if the marginal alignment probabilities for each read are the same.
Finally, there are other unexpected benefits that you can get from paired-end sequencing. All of the above applies to just the mapping step, but applications like structural variant detection and chromosome conformation capture really benefit from paired-end sequencing.
If you have single-end reads for structural variant detection, then you will only detect breakpoints if the read overlaps a breakpoint. But if you have paired-end, you not only have twice the amount of DNA sequenced to look for breakpoints, you can also estimate if there is a breakpoint between the two reads (in the inner distance from your diagram) if the two reads align really far apart from each other, or on separate chromosomes. The idea is similar for chromosome conformation capture and the ligation junction sites.
In paired-end sequencing, you have an expected fragment length. So if Read 1 maps to two different places, whereas Read 2 only maps to one place. Then you can determine which one is more likely for Read 1 to map to. It cannot be too far away from Read 2 given the expected fragment length.
It is not necessarily true that paired-end sequencing gets twice the amount of data. The total amount of data can be determined when running sequencing. If single-end has 20M reads, I can just stop sequencing at 20M as well for paired-end.