Interpretation of MM:Z:A+a? tag from m6A basecaller model

[sarah-ku]

Hope you don't mind us asking here about the output from the m6A methylation model.

We converted our DRS (rna004 kit) data from fast5 to pod5 and tested out the command:

dorado basecaller sup,m6A_DRACH data.pod5 > output.bam

here is an example of one of the entries from the resulting BAM file:

414b0961-3412-4660-a4f1-118de6d22719 4 * 0 0 * * 0 0 GGCGAGCAGGGAGGCAAAGCTCGCGCCAAGGCCAAGACCCGCTCTTCTCGGGCCGGGCTCAGTTTCCCGTGGGGGCCGAGTGCATCGCCTGCTCCGCAAAGGCAACTGCACGGCGGAGCGGGTGCTGGAGCTCCGGTGTCCCTGGCGGCGGTGCTGGAGTACCTGACCATCGAGATCCTGGAGCTGGCTGGCAACGCGGCCGCGACAACAAGAAGAATTCGTCATCATCCCCGCGCACCTCGAGCTGGCCATCCGCAACGATGAGGAGCTCAACAAGCTTCTGGGCAAGTCATACATGGTGGCGTCCTGCCCAACATCCAGGCCGTGCTACTGCCCAAGAAGACCGAGAGCCAAGGCGGGCAAGTAGAAGCCTGGATTAGTTTGCAGCAACTCAATCCCAAGGAACCAAAGGCTCAGAGCCTTGGGGTGGCCCCAGCCCCCACCCCCGCCCTACAACTTATCAGCCCATATCAACCCTGCCCCCTCCCCCTCGCCCCCTCGCCCTCTCAAAACACCCC ((((???==>;<9101869844343196=A?AB67778@+*)(*)(),/111966.(($$$%),.0)))(((+121113447889=;5;85555:=@A3=?9481&&%%%%%(-''())6=>>=>@C**)+.000//+,2;9CC98;;<?6).>JA:?87753:*))))*5////389;<567:66678;:@=@;94&%(34>99889888.%$$$$%&&&+11676(.75.'''''''''(738=73353=?CAAC99858:6<@:9?BED=7>6876769<<42'$####$%((,11123,+..07<<39<==76444//5775440231165<<:=8>6FGB884-*+**(%&(+/035666400011>>?@=>=<<;;::8//.../))-/,1((3379433-45*'&'(+66)('()2*$$#""###$$%&&%&&'*($%&&%%&%')--+++)())('**),+%%('$##$$%')&&&)'(&$###$&$##$#$%)%$###"##$$%')'%$ qs:i:9 du:f:5.71875 ns:i:22875 ts:i:0 mx:i:3 ch:i:88 st:Z:2024-02-27T06:33:59.408+00:00 rn:i:31356 fn:Z:output6.pod5 sm:f:79.4189 sd:f:19.3945 sv:Z:quantile dx:i:0 RG:Z:c884b8754e91b8b445f4f47f572cedd4a0678cca_rna004_130bps_sup@v3.0.1 MN:i:518 MM:Z:A+a?,10,13,43,19,19; ML:B:C,0,1,1,1,199

My questions are:

• does MM:Z:A+a?,10,13,43,19,19 refer to the detected m6A site positions (in 1-based index form) and ML:B:C,0,1,1,1,199 the associated confidence scores?
• if our above interpretation is correct, does this imply something has gone wrong since these positions do not refer to "A" nucleotides (e.g. position 19 with the 199 score is a G)
• Also if ML:B:C are scores why are there positions with scores of 0? And why is 19 listed twice with two different scores?
• In terms of clearing up probable misunderstandings on our part - is there detailed documentation somewhere which explains how to interpret the output of the modification models? We looked both here and on the community pages could not find anything relevant :).

We are using the required rna004 kit for this model and running on converted pod5 files on a GPU server.

[HalfPhoton]

Here's a link to the SAM tags documentation

The definition of MM numeric values - emphasis on to skip

... comma separated list of how many seq bases of the stated base type to skip, stored as a delta to the last and starting with 0 as the first (or next) base, ...

MM:Z:A+a?,10 - skips 10 "fundamental" A bases - 19 appears twice out of coincidence here.

The definition of '?':

When this flag is ‘?’ there is no information about the modification status of the skipped bases provided.

Kind regards, Rich

[ArtRand]

Hello @sarah-ku,

The MM and ML tags are difficult to interpret by eye. I recommend using modkit extract (docs) to transform the MM/ML tags into a table. @HalfPhoton is correct about the MM tags being the number of "skips" not the actual positions, the ML scores are 0-255 probability bins, so 0 is the lowest probability of modification and it's reasonable for multiple positions to have the same prediction probability.