3
$\begingroup$

In case of Drop-seq, we have paired end data.

Read 1: Cell code + UMI (unique molecule identifier)

Read 2: The transcript information

But I have a problem/doubt with the sample I am working on.

The sample I am using is the following (Check the "Reads" tab).

As you know the Drop-seq is "paired-end", we are expected to see two reads for a spot. Although this sample say paired-end, it has only one read per spot.

For example I can share a link of a different scRNA-seq data where you can properly see two reads for a spot

Example sample, you need to check the "Reads" tab.

Where I am going wrong?

I asked one of the main authors of the paper. The following is the reply I got :

I recommend that you download the aligned BAM files that are hosted in the same GEO record. Read 1 is already processed into the cell and UMI barcodes and held as custom tags (XC and XM) in the BAM files. The cells are already barcode-corrected, so if you use those files, your cell barcodes will line up with mine; if you start from FASTQs, they will not. For most aligners, you can just use the BAM file as input to realign. (It has all reads, even those that did not align.)

But I could not find any "XM" or "XC" keywords in the bam file :(

To understand his reply you have to be familiar the processing steps of the Drop-seq.

Looks like they have submitted some kind of processed data. I could not figure out how much the data is processed. I am trying to use the data starting at different processing steps but I could not figure out how much the data is processed.

$\endgroup$
2
  • 1
    $\begingroup$ Where did you even find their BAM files in GEO? I only see the output from CellRanger (UMI counts and such). $\endgroup$ – Devon Ryan Nov 13 '18 at 9:43
  • 1
    $\begingroup$ My guess is the authors uploaded BAM files to GEO which have been converted back to fastq in SRA, losing all the extra annotation. $\endgroup$ – Ian Sudbery Nov 13 '18 at 10:28
2
$\begingroup$

I received the following message from the author of another Drop-seq paper:

After much conversation with the folks at the SRA, it appears the SRA compression algorithm automatically deletes certain subsets of bam tags and does so in cryptic way (in the sense of not alerting the uploader). Obviously this is going to be an issue for many single cell datasets encoded this way, and I hope the SRA will eventually change their system to accommodate.

If you look at the "Download" tab of the same page, there now exists a link to download the originally uploaded BAM file.

Alternately, you can use the SRA AWS tutorial to download the files on that platform.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy