[aniruddha.otago]

yeah when we were performing the analysis the latest version was 1.2 ( release, April, 2011). every program has newer version each week, this analysis was done at that time with latest versions of all the aligners at that particular time.

Cheers,

Aniruddha.

[aniruddha.otago]

Hi Yuanxin,

Thanks, we have solved the problem and it came up with reasonable percentage of methylation. It was useful for the analysis. Thanks.

[aniruddha.otago]

Hi rskr,

We never really thought of trying to do that. We were attempting
to see the performance with real RRBS data which we did. I'm not
sure that going to the trouble of trying to generate random CpG fragments
would have achieved very much at all. Any attempt to 'randomise' the
C/T conversions of somehow randomly chosen CpGs can't be guaranteed
to reflect real life methylation patterns. Until the rules for positional
methylation are more fully understood, attempts to generate randomised
data are likely to be influenced by artefacts of the process used.

[rskr]

Hi rskr,

We never really thought of trying to do that. We were attempting
to see the performance with real RRBS data which we did. I'm not
sure that going to the trouble of trying to generate random CpG fragments
would have achieved very much at all. Any attempt to 'randomise' the
C/T conversions of somehow randomly chosen CpGs can't be guaranteed
to reflect real life methylation patterns. Until the rules for positional
methylation are more fully understood, attempts to generate randomised
data are likely to be influenced by artefacts of the process used.

It may true that the distribution of CpG methylation is potentially not random. However simply calculating unique matches doesn't tell you anything about the performance of the mapper, which is what you claimed to have measured. It may very well be that the correct mapping was indeed ambiguous, and you don't know. If you can't model the CpG methylation, then you can't claim to have measured the performance of the mapper.

[aniruddha.otago]

It may true that the distribution of CpG methylation is potentially not random. However simply calculating unique matches doesn't tell you anything about the performance of the mapper, which is what you claimed to have measured. It may very well be that the correct mapping was indeed ambiguous, and you don't know. If you can't model the CpG methylation, then you can't claim to have measured the performance of the mapper.

it is done on a reduced represented genome. so, say there is a fragment of 100 bp and 8 reads uniquely mapped to that. mind you each fragment and read has to have a CGG or TGG (MspI site) at the start to be able to map. say, these 8 reads has total 16 CpG site. so, we counted this as a no of unique aligned CpG site. we did make commnet with a figure that even for the same region different programs showed different mapping. However, randomising CpG site didnt seem appropriate to us in the context of this paper. "It may very well be that the correct mapping was indeed ambiguous, and you don't know". If thats the case then no point doing any other step, because simply it is not giving you correct information. The aligners choosen here are widely used in the field and are well established to yield reliable results and we have no reason to believe that the correct mapping is ambigious.
we wanted to give a feel to the molecular biologists that if you have a RRBS dataset and use this widely used aligners, this is what you can get and what are the aspects one should be aware of. and also, a biologists is more likely to will work with real sample to analyse, not on "randomised CpG data".

[rskr]

... we wanted to give a feel to the molecular biologists that if you have a RRBS dataset and use this widely used aligners, this is what you can get and what are the aspects one should be aware of. and also, a biologists is more likely to will work with real sample to analyse, not on "randomised CpG data".

That's fine except I gather that your data is already out of date in technology terms(GAIIx), and wasn't very good to begin with, since you ended up trimming to make the mapping rates go up, indicating the quality dropped severely near the ends. Is that because of the RRBS protocol or something else? And as you mentioned you really don't have a model for methylation so there isn't any reason to believe that your results would be reproducible.