Hello Modkit developers,
I would like to ask for guidance on comparing global 6mA levels between two Nanopore samples with unequal sequencing coverage and no biological replicates.
Both samples were processed using the same Dorado model, Q15 filtering, reference genome and Modkit settings. A fixed adenine probability threshold of A:0.9394531 was used for both samples to avoid differences caused by dynamic threshold estimation.
I calculated the global 6mA percentage as:
100 × sum(Nmod) / sum(Nvalid)
where:
Nvalid = Nmod + Ncanonical
The results were:
-
barcode04, complete dataset
- Mean genomic coverage: 203.425×
- Mean valid coverage per adenine site: 91.639
- Adenine sites with
Nvalid >10: 17,117,572
- Global 6mA: 0.99147%
-
barcode04, subsampled to approximately 52×
- Mean valid coverage per adenine site: 24.584
- Adenine sites with
Nvalid >10: 16,683,017
- Global 6mA: 0.99106%
-
barcode07
- Mean genomic coverage: 51.723×
- Mean valid coverage per adenine site: 23.665
- Adenine sites with
Nvalid >10: 16,696,954
- Global 6mA: 0.94674%
Subsampling barcode04 to approximately the coverage of barcode07 substantially reduced its mean valid coverage per adenine site, as expected. However, it retained approximately 97.46% of the sites that originally had Nvalid >10, and its global 6mA estimate changed only from 0.99147% to 0.99106%.
My questions are:
-
Is sum(Nmod) / sum(Nvalid) an appropriate genome-wide comparison when samples have substantially different coverage? Does the stability observed after subsampling indicate that coverage is not materially biasing the global 6mA estimate?
-
For this global comparison, would you recommend using all available reads or subsampling the higher-coverage sample to match the lower-coverage sample?
-
How reliable is a one-versus-one comparison without biological replicates? Should it only be interpreted as a descriptive comparison between these two samples, and are biological replicates required for robust single-site or DMR inference?
Thank you very much for your help.
Hello Modkit developers,
I would like to ask for guidance on comparing global 6mA levels between two Nanopore samples with unequal sequencing coverage and no biological replicates.
Both samples were processed using the same Dorado model, Q15 filtering, reference genome and Modkit settings. A fixed adenine probability threshold of
A:0.9394531was used for both samples to avoid differences caused by dynamic threshold estimation.I calculated the global 6mA percentage as:
100 × sum(Nmod) / sum(Nvalid)where:
Nvalid = Nmod + NcanonicalThe results were:
barcode04, complete dataset
Nvalid >10: 17,117,572barcode04, subsampled to approximately 52×
Nvalid >10: 16,683,017barcode07
Nvalid >10: 16,696,954Subsampling barcode04 to approximately the coverage of barcode07 substantially reduced its mean valid coverage per adenine site, as expected. However, it retained approximately 97.46% of the sites that originally had
Nvalid >10, and its global 6mA estimate changed only from 0.99147% to 0.99106%.My questions are:
Is
sum(Nmod) / sum(Nvalid)an appropriate genome-wide comparison when samples have substantially different coverage? Does the stability observed after subsampling indicate that coverage is not materially biasing the global 6mA estimate?For this global comparison, would you recommend using all available reads or subsampling the higher-coverage sample to match the lower-coverage sample?
How reliable is a one-versus-one comparison without biological replicates? Should it only be interpreted as a descriptive comparison between these two samples, and are biological replicates required for robust single-site or DMR inference?
Thank you very much for your help.