A Modification-free Negative Control Defines the Operating Points at Which 5mC and 6mA Can Be Called from R9.4.1 Nanopore Data

Abstract

Nanopore sequencing reads DNA base modifications from the ionic-current signal, but the R9.4.1 chemistry has no vendor model for N6-methyladenine (6mA), and third-party models are applied without an absolute error estimate. We measured the false-positive rate of two 6mA callers and of one all-context 5-methylcytosine (5mC) caller against a negative control that cannot carry either mark: PCR-amplified cDNA, in three independent controls from two cell lines. False-positive rates were resolved per 5-mer context and compared with the specificity that each modification’s prevalence demands. For CpG methylation, the measured false-positive rate remained approximately two orders of magnitude below the maximum rate permitted under the stated assumptions, and no context restriction was required. For 6mA, the pooled false-positive rate at the commonly used probability threshold of 0.7 was 23.2-fold higher than the maximum admissible rate. Most of the excess background was contributed by 30 of the 256 adenine-centred contexts, none of which carries a cytosine immediately 5′ of the called base; excluding them improves specificity 5.9-fold at threshold 0.95 and brings the caller inside the requirement there. The rank ordering of context-specific false-positive rates was correlated between cell lines (Spearman ρ = 0.862), although absolute rates were not identical. Under the tested Dorado/NEMO configuration, a second 6mA caller classified 0.63% of adenines as modified in amplified DNA even at threshold 0.99, corresponding to 277 times the admissible false-positive rate and insufficient specificity for rare 6mA detection. Genomic DNA prepared with the FFPE repair step omitted, so that 8-oxo-dG is preserved, inflates apparent 6mA 9- to 15-fold while leaving CpG calls unaffected.