Should poorly sequenced samples be excluded?

[Jane M]

Dear all,

I am wondering if a sample should be excluded when it is very poorly covered.
Can you please tell me what do you think and what do you do with such samples?

In my case, I have RNA-Seq paired-ends data for 3 conditions: Control (4 biological replicates), Leukemic cells with mutation in gene X (6 biological replicates) and Leukemic cells without mutation in gene X (4 biological replicates).

1 out 4 control samples has 5986853 raw reads per fastq. 78.3% of alignment rate. 3.2M of reads aligned on the transcriptome.

Since I only have 4 samples for the control condition, I did not delete this sample. (I know that a lot of studies are done with only 3 samples per condition but it is probably not enough to get robust results). Do you think such a sample can help or should be removed?

Thank you,
Jane

[westerman]

My bottom line gut feeling is that using 3 control samples should be fine but I would be hesitant to toss out the 4th sample without knowing why it did not map at a high rate. I presume that the other control samples mapped at a much higher rate. Was there any technical reason that the mapping was worse with the poor sample? If it was because of poor quality reads then throwing away the sample is probably best (and just use the 3 good samples). If the reason was because of contamination of the sample then I think you could clean the sample of the contamination and remap. But it also depends on how many reads your other samples have -- you could be getting too few reads to make a good sample. Maybe there is another reason for the poorer mapping?

[Jane M]

Thank you for your answer.

Indeed, the other controls mapped at higher rate: 85.2%, 89.1 and 90.7%. The other samples mapped at 88-92%.

I don't know why the mapping is worse for this sample. This is probably related to the small amount data that I got from the beginning: 1.5Go per fastq file compared to 7-37Go for the others.
I did the mapping on filtered (by Trimmomatic) reads with the same parameters for all samples. I did not filter a lot of reads for this control: from 5986853 to 5888867 for R1 and 4901412 for R2.

For the other samples, I have between 25-100M, but one at 16M of reads that aligned to the transcriptome.

[TiborNagy]

FastQC can show some information about the quality of the sequences.

[Jane M]

FastQC can show some information about the quality of the sequences.

I use FastQC to check the quality 2, before and after Trimmomatic.
There is nothing special for this sample, as shown in the attached files. I think the problem comes from the library.

Can you please tell me what you do with such a sample?
Jane