Dorado (by modkit dmr) gives different results between replicats

[DelphIONe]

2 replicats sequenced with different version of MinKnow but the same dorado version used gives some extra 6mA not legitim

Rep1 sequenced on May, Rep2 sequenced on November. Dorado (dorado-0.7.2-linux-x64) and modkit (0.3.1) used for both.
IVT sample used for modkit dmr has been sequenced in the same time as rep1.
Rep2 with modkit dmr gives new (different) positions compared to rep1 and these positions can't be true.

Steps to reproduce the issue:

~/softs/dorado-0.7.2-linux-x64/bin/dorado basecaller [email protected] pod5/ --modified-bases m6A > sample_6mA.bam
modkit pileup sample.bam sample.bed --filter-threshold A:0.8 --mod-thresholds a:0.9 --ref file.fa --log-filepath sample.log --with-header --edge-filter 20

I obtain strange results for rep2. I'm trying to understand where the problem might come from. Maybe that modkit sample_probs is the key but I'm not sure ? I have attached the 3 probabilities.txt files.
probabilities_rep2.txt
probabilities_rep1.txt
probabilities_IVT.txt

The only one difference between rep1 and rep2 is the version of MinKNOW. 24.02 for rep1 and IVT and 24.06 for rep2. Do you think this is the reason for the difference? If I resequence IVT sample with the same MinKNOW version (as rep2), my problem will be resolved ?

Can you help me please ?

Many thanks

[ArtRand]

Hello @DelphIONe,

Rep2 with modkit dmr gives new (different) positions compared to rep1 and these positions can't be true.

Could you elaborate on what you mean by "positions can't be true"?

[DelphIONe]

Hello @ArtRand

Thanks for your reply.
The detected position until now with rep1 are potentially real 6mA because we fall in DRACH motif (or almost by 1 base)
The new (and artefactual) positions are not DRACH motifs, but rather a CAG pattern very often.
I don't understand, as if it were a model problem, but the model used in the 2 cases is the same.
For these new positions in rep2, compared to IVT, there was more signal in the IVT compared to rep1 and it's the opposite for rep2 (more signal in rep2 compared to IVT).
In parallel communication with Nanopore, I learned that the sequencing temperature between the 2 versions of MinKNOW used was altered by 3 degrees. Could this not explain the difference in the modkit sample_probs probabilities files ?