A question on Illumina paired-end reads alignment. Merging from different samples

[linudz]

Hi folks, I have a question on how to better merge paired-end reads coming from different samples. Basically, we do align genome resequencing fastq with bwa mem, and compress the resulting .sam file into a .bam.

If we have to merge two different samples, there are basically two ways. The first is to merge the fastqs and align the resulting file, the second is to make use of samtools merge to merge the .bam files.

My concern is whether the two procedures are equally valid, or there is some relevant difference in the outcome.

I think that everything revolves around the functioning of the Burrows Wheeler Transform Alignment. I have broadly understood the application of the BWT and of the indexing, but I still wonder if the number of reads affects the results of the alignment, or each read is aligned independently.

Can anyone give me more insights on this?

[GenoMax]

Hi folks, I have a question on how to better merge paired-end reads coming from different samples. Basically, we do align genome resequencing fastq with bwa mem, and compress the resulting .sam file into a .bam.

If we have to merge two different samples, there are basically two ways. The first is to merge the fastqs and align the resulting file, the second is to make use of samtools merge to merge the .bam files.

My concern is whether the two procedures are equally valid, or there is some relevant difference in the outcome.

Either way should be fine after you take the "note" below into account.

Note: When you are referencing "merging samples" are you referring to technical replicates of the same sample? Merging would be appropriate only in that case. If "samples" are true biological replicates then you would want to keep them separate for downstream analysis.

but I still wonder if the number of reads affects the results of the alignment, or each read is aligned independently.

Can anyone give me more insights on this?

Each read pair is independently aligned to the reference so there is no effect of the amount of data on actual alignments.

[linudz]

Thank you very much for your answer. This is going to be done on replicates of the same sample, so it should apply. Thanks again

[GenoMax]

Thank you very much for your answer. This is going to be done on replicates of the same sample, so it should apply. Thanks again

Just want to re-emphasize. Are these technical replicates or biological? Merging data is only appropriate for technical replicates.

[Brian Bushnell]

Because technical replicates could have different insert sizes, and the average (or estimated) insert size can have an effect on alignment, you may get different results even for technical replicates between merging prior to mapping versus merging after mapping. In many cases this difference is trivial and in general mapping the reads independently is not guaranteed to give you an optimal answer with respect to insert size anyway, since there is an order dependency, but it's something to be aware of.