ONT sequencing methylation data normalisation

[Sarrus]

I have methylation data from ONT sequencing expressed as percentage of methylation and/or number of reads. I don't know if I'm supposed to normalize the data and how to handle replicates. Some ideas: I have data in triplicate. I want to normalize the data (quantile normalization is the preferred option) and/or collapse data to have 1 file/dataset from 3 replicates (doing weighted average maybe). So 1 idea would be just to perform weighted average of each methylation value doing [%M*NR] where %M is the percentage of methylated reads and NR is number of reads, per each methylated position. In this case I will lose the meaning of the percentage value in the downstream analyses because I will have a weighted value. otherwise I can perform quantile normalization on the number of reads (total and methylated) and then calculate percentage of methylated reads. I would like the opinion of someone with a better statistical background than me thanks for your kind help!

%M-S1 NR-S1 %M-S2 NR-S2 %M-S3 NR-S3
Meth1 20 60 15 54 41 12
Meth2 40 14 78 52 13 65
Meth3 12 94 73 19 37 70
Meth4 36 77 69 14 26 74

[BasesBreaker]

Okay, so here's my take on this. Since you have triplicates and want to both normalize the data and condense it, this is my recommendation.

Start by normalizing the number of reads (both total and methylated) across your replicates using quantile normalization. In this case, you'll ensure that all your samples have a similar distribution, which should make it easier to compare them. Now after normalization, you should recalculate the percentage of methylated reads for each position.

And if you want to collapse the data from 3 replicates into 1 dataset, then a weighted average makes the most sense to me. For each position, you can calculate the weighted methylation percentage using the formula you've included in your post. After you get this weighted value for each replicate, average these values for the three replicates. This will end up giving you a single value that takes into account both the methylation percentage and the number of reads.

If you do all of this, you will normalize the data and then condense it into a single dataset that's more representative of your three replicates.

I'm a bit rusty at this type of work so it wouldn't hurt to get a second opinion, but based on what I remember, that's what I would do.