GC bias

[Richard Finney]

Running a qualitative regression (i.e. regression using qualitative variables, not quantitative variables) would be interesting.

For random locations, break down the surrounding reference to into components.

For example a location may have the surrounding sequence

ACTGAACCTTGGTAAACCCTTTGGG (note solo 'T' in the middle)

The equation would be
COVERAGE = ACTGA + CTGAA + TGAAC + GAACC + ... + TTGGG
(note the sliding window of size 5. different windows sizes could be used)

Let's say the coverage is 40 so, youl'd set up your data input file like this ...
40 ACTGA CTGAA TGAAC GAACC ... TTGGG
.... more data ...

and , using R, run a glm().

This might indicate if there are certain sub-sequences that contribute to variable coverage.

A variation on a theme would be using sequences from the reads and compare using sequences from the reference.
There might also be filtering locations based on "mapability".

[shuteo]

Hi,

I am a statistician rather than geneticist/biologist so would really be grateful if someone can explain the cause/origin of GC-content bias with sequencing coverage. Many studies have observed a unimodal relationship where coverage decreases at high AT or high GC.
From what I understand, since AT bonds are weaker than GC bonds, in the PCR step, fragments with extreme GC (strong bonds) may not denature completely to form the single stranded DNA, hence we see a trend of decreasing coverage as GC increases.
But what about the decreasing coverage in regions of extreme low AT?
Can anyone explain?

[colindaven]

We think the stronger binding of GC pairs are the cause of the problem. I can't imagine that all of the bases we've observed to be uncovered are due to DNA binding elements.

Some samples only got thousands of reads, as opposed to the millions expected. We'll be trying out modified PCR protocols and reagents to try to correct this.

The inverse correlation between increasing GC content and coverage even occurs within highly conserved genes, most of which I wouldn't expect to be a target for sigma factors in bacteria.


This paper from the Broad institute shows some nice solutions for getting better coverage at high GC.

Aird, D.; Ross, M. G.; Chen, W.-S.; Danielsson, M.; Fennell, T.; Russ, C.; Jaffe, D. B.; Nusbaum, C. & Gnirke, A. Analyzing and minimizing PCR amplification bias in Illumina sequencing libraries. Genome Biol, Genome Sequencing and Analysis Program, Broad Institute of MIT and Harvard, 320 Charles Street, Cambridge, MA 02141, USA., 2011, 12, R18

Abstract: Despite the ever-increasing output of Illumina sequencing data, loci with extreme base compositions are often under-represented or absent. To evaluate sources of base-composition bias, we traced genomic sequences ranging from 6% to 90% GC through the process by quantitative PCR. We identified PCR during library preparation as a principal source of bias and optimized the conditions. Our improved protocol significantly reduces amplification bias and minimizes the previously severe effects of PCR instrument and temperature ramp rate.

[Richard Finney]

Curiosity ...
Is it GC content the cause of the difficulty or co-incident with the problem?

Are the actual "[G,C]-[G,C]" bonds interfering with the measurements?
Or are there histones(?) or transcription factors(?) binding the DNA at high "GC" and these molecules can't be cleaned off of the DNA and therefore prevent just having the DNA isolated?

[Zaag]

The Platinum PCR mix from SOLiD is not good with GC-regions.

[colindaven]

Hello all,

it would be really interesting to find out what people have sequenced on their SOLiDs, and if it worked well with the standard DNA seq. protocols.

To start off, we have done:

-human exomes (worked fine)
-human genomes (fine)
-high GC bacterial genomes (62 and 66% - not so good)

We have found coverage decreases strongly with GC at high GC contents.

Has anyone else seen this, or observed that the main protocols are optimised for ~40% GC eukaryotic resequencing?
(Solid 5500xl, single end fragment library)