Hi everyone, I have a question regarding the analysis of data derived from modkit. I am working with two sets of data (for the moment, a few samples): one concerning samples sequenced without PCR amplification using MinION r10, and another set of data derived from PCR and then sequenced using the same technology. Some samples are the same. I was wondering if there is a method (statistical or a boinformatic pipeline , etc.) that would allow me to determine whether the methylation values obtained constitute background noise on the PCR samples, as I do not expect significant methylation on these. Additionally, if there is a way to establish a threshold for background noise. I have read a lot about it, but perhaps you can help me. Thank you all.
I don't know about a threshold for background noise in methylation, but I don't think this is a reasonable estimation of background methylation noise. Methylation can only be estimated from PCR-free runs, as PCR should remove any post-translational modifications on your reads. So, to me what you would be estimating from your outline is how many spurious methylations are introduced or due to inference errors. This is different to estimating background methylation.
Thank you very much for your response. I've been calling it background noise, but perhaps we're saying the same thing. I mean, I really want to see if I get methylation percentages in PCR samples that deviate from zero and thus disturb the analysis. Since nanopore sequencing is sensitive to the electrical passage of bases, I want to see if this can affect it. So, the first step of the analysis is to look for any biases due to the technology, and in a second step, conduct DMR analysis.
I really like the idea of trying to quantify background noise. However, PCR is not used to process methylation calls reads. So by using PCR and then quantifying numbers of spurious methylations, you're quantifying noise from a process that isn't involved in most cases. That make sense? Essentially, to me, you're asking an interesting question about a technique that isn't involved.
I think a more meaningful approach would be taking many technical replicates of a number of samples, and quantifying variation among replicates. I suspect this is often done, but seldom reported. Here, you'd be quantifying background noise based on your target species and the reaction and sequencing kits.