more 'C's than 'G's!!

[Bill Amos]

Dear Forum,

Does anyone have information about a G vs C bias? I am analysing 1000 genomes data based on raw fastq files ~5Gb in size. For other reasons I counted the occurrence of each base and found, to my surprise, that 'C's outnumber 'G's by about 10% (though the size of the bias varies between sequencing centres). To guard against artefacts, I then counted G-C frequencies for each base position in each read (I only look at forward runs, so have bases 1 to 100). I see that the G vs C bias varies with base number, being very variable for the first and last 5 bases and otherwise generally rising (= more 'C's) along the read.

Is this a well-known phenomenon? Any thoughts gratefully received!

Cheers
Bill

[Brian Bushnell]

Sounds like a calibration problem. I have not seen this in our data.

The variability at the ends, though, I have seen. This is sometimes due to adapter contamination, and sometimes due to the fact that base-calling uses information about adjacent bases, which is absent at the tips.

[Bill Amos]

have you looked specifically at C / G imbalance?

Dear Brian,

Thanks. The pattern seems remarkably pervasive - I have looked at data from SC, BGI and BCM (all show quite strongly in ~75 reads, each 5Gb) and WUGSC (less strong effect). There does also seem somewhat of a population effect, though I'd need to look at more samples to be sure. If you take an index 2*(C - G) / (C + G), where C and G are the numbers of 'C's and 'G's, this typically rises from 0.08 to o.12 between based 5 and 95.

Cheers
Bill

[dariober]

I then counted G-C frequencies for each base position in each read (I only look at forward runs, so have bases 1 to 100).

Hi- Isn't this the information reported by FastQC, module "Per base sequence content"?

I looked at the FastQC report for some data we have (ChIP-Seq, FAIRE-Seq) and I don't see much difference between C and G after the first 5-10bp. Although the first bases are indeed very biased but I think this is due to the non-so-random fragmentation.

[Bill Amos]

FastQC

Yes, FastQC does present the relevant data, though not as a ratio / difference. Slight (G 5% down, C 5% up) might not show up very clearly, either in this plot or, of course, as GC content, particularly if the first few bases are wayward. I guess this sort of bias 'comes out in the wash' when the reads are aligned, but I'm trying to analyse data without any of the biases that alignment can introduce!