dropSeqPipe configuration

[grst]

Hi @Hoohm,

while creating the sample.csv files, I came up with a couple of questions:

• Is the n_cels parameter in samples.csv mandatory? I plan to filter downstram with scanpy, and cannot find the 'expected cells' per library in the description of every dataset.

• Read length only refers to R1 read length, right?

• Some samples consist of multiple runs with different read lengths each (:roll_eyes:). How to deal with them? Suggestion: treat them as individual samples

• Which brings me to the next question: does it make any difference if I (a) concatenate two fastq files and treat them as a single sample or (b) leave them as they are and treat them as two samples.

[Hoohm]

• Yes, this you have to provide. It will use it as the number of cells to extract using umi-tools. If you have no idea at all, we shoule use a very high number. Hopefully with the sparse matrix, it should pose a big problem
• Read length of R2, the RNA. This is used for index generation os STAR.
• Exactly the same samples with different read lengths? That is odd. I would keep them apart and combine them at the very end based on the cell barcode.
• You can concatenate different lanes together, this is usually the case with 10x data. Although I would not concatenate samples that have different lengths. Trimming and error rate are gonna be biased.

[Hoohm]

For the first question, 10x provides a whitelist of all possible cell barcodes. We can use those if you want. meaning we extract all 737k cells and then filter.

[grst]

For the first question, 10x provides a whitelist of all possible cell barcodes. We can use those if you want. meaning we extract all 737k cells and then filter.

That's basically what I do now for the lambrechts data from cellranger. It seems to work quite well. But how would that look like for the other protocols? Or would it work if I just specify a large, arbitrary number, say 200,000?

• Exactly the same samples with different read lengths? That is odd. I would keep them apart and combine them at the very end based on the cell barcode.

They have the same GSM identifier on GEO, but multiple SRR identifiers on SRA. I think they contain different cells. -> keeping them seperate defintely makes sense.

[grst]

They actually have different read lengths within the same file. No idea how they got there. It's all Illumina HiSeq 2500. >zless SRRXXXXXX_2.fastq.gz

What will STAR do when it does not have an index for a certain read length? Actually, the authors use STAR, too, but I couldn't find out so far how they generate the index.