Tools to perform QC of methylation calling

[ynwh]

Hi,

My BS-seq mapping generated different methylation level value (count_C/(count_C+count_T)) at different bases. Some values are very low (~0.07). I wonder if there is a widely accepted threshold to filter unreliable methylation calling or any tool can do this?

Many thanks

[PeteH]

To clarify, are you wanting to identify cytosines that show evidence of methylation and thus you want to filter out any with low estimated levels of methylation that you think might be simply attributable to 'error' (be that sequencing error, bisuflite-conversion error, mapping error, base calling error, etc.)?

If so, you might have a read of the supplementary material of Lister et al. (Nature, 2009 and 2011: http://www.ncbi.nlm.nih.gov/pubmed/19829295, http://www.ncbi.nlm.nih.gov/pubmed/21289626). They use a binomial test to "[establish] the minimum threshold number of cytosines sequenced at each reference cytosine position at which the position could be called as methylated, so that out of all methylcytosines identified no more than 1% would be due to the error rate."

I don't know of any software off-hand that implements this test for BS-seq data but it is a simple binomial test that could be implemented in, for example, R.

I would be reluctant to use a simple threshold (e.g. 0.3) because the estimated methylation level is a function of the coverage in BS-seq, as is the standard error of your estimate. And I don't think there is a "widely accepted" threshold. That said, this paper (http://www.ncbi.nlm.nih.gov/pubmed/23598999) does provide thresholds, although the paper they cite is a study using microarrays (where thresholds may be more justified) rather than BS-seq.

Hope this helps,
Pete

[ynwh]

Pete,

Many thanks for your answer. That's really helpful.

[chadn737]

You really need a control for you BS treatment, an unmethylated genome which was also treated with BS so that you can estimate the conversion. In plants the chloroplast genome serves as a convenient internal control, otherwise you should spike in lambda phage in your samples. That is essential to the calculations used by Lister et al.