Merged vs unmerged fastqs

[harar]

I have several samples each in their own folder. Paired end. Each sample's folder looks something like this in the following format:


****_L006_R1_001.fastq.gz and ******_L006_R2_001.fastq.gz
****_L006_R1_002.fastq.gz and ******_L006_R2_002.fastq.gz.
...
.
.
.
****_L006_R1_006.fastq.gz and ******_L006_R2_006.fastq.gz.
****_L007_R1_001.fastq.gz and ******_L007_R2_001.fastq.gz.
.
.
.
****_L007_R1_006.fastq.gz and ******_L007_R2_006.fastq.gz.

What I've done so far is merged all the R1s (irrespective of lanes), and all R2s (again, irrespective of lanes). And have been aligning them, and calling fusions on the merged R1.

How do I interpret any fusion results of ****_L006_R1_002.fastq.gz and ******_L006_R2_002.fastq.gz. VERSUS R1-merged.fastq.gz vs R2-merged.fastq.gz? Does R1-merged/R2-merged represent the entire sample properly or should I call fusion on each R1,R2 and find some sort of average?

This also applies to finding indels/cnvs/etc, do you guys usually call it on the merged fastq file or each unmerged one, and then find an average?

thank you!

[GenoMax]

You can merge R1 (and R2) reads for a sample (use the same order for R1/R2 files)(samples are part of a pool that ran on multiple lanes, I presume) that ran on multiple lanes. In general people may find it easier to process the file pieces in parallel until the alignment step and then merge the resulting BAM files before sorting/indexing the final result. You can do the fusion detection on final file.

[harar]

Thank you!

Fusion detection has caveat that it requires fastq files. Are these R1-merged.fastq.gz and R2-merged.fastq.gz complete representatives of the sample? I ask because as an example, I ran fusion detection on a individual pair of _R1_001 and _R2_001, and counted 6 junction reads for a certain gene. Then did it for the R1-merged and R2-merged and got fewer fusion detections.

Any idea why? Are these individual fastqs (unmerged fastqs) representative of the entire sample?

[GenoMax]

Merged fastq files should be identical representations of individual pieces unless whatever fusion program you used had trouble completely reading the merged files. Did you check to make sure the reported reads processed were identical in both cases. How did you merge the files (by cat?)

Illumina's data processing software (bcl2fastq) actually has an option (--no-lane-splitting) that will give you a single file per samples for samples that may have run across multiple lanes of a flowcell as a part of a pool.