nanopolish: Error when trying to variant call using an R10 model

Hi Jared,

I have trained a basic k-mer model for R10.4.1 using your R10 branch in nanopolish, using the recent R10.4.1 base models provided by ONT. I am trying to evaluate some variant calls using that, but I am getting the following error

nanopolish variants -r reads.fastq -b reads.bam -g ~/scratch/hg38noAlt/hg38noAlt.fa -p2 --models-fofn=a.txt
Error: unknown model:

The contents of a.txt looks like:

../r10-models/r10.4.1_400bps.nucleotide.9mer.model

The head of the …/r10-models/r10.4.1_400bps.nucleotide.9mer.model is like:

#model_name     r9.4_450bps.nucleotide.6mer.template.model
#kit    r9.4_450bps
#k      9
#strand template
#alphabet       nucleotide
AAAAAAAAA       54.20490641     3.0     1.0     1.0
AAAAAAAAC       58.59079727     3.0     1.0     1.0
AAAAAAAAG       55.952068       3.0     1.0     1.0
AAAAAAAAT       58.43352009     3.0     1.0     1.0
AAAAAAACA       63.65995928     3.0     1.0     1.0

I initially had the following but changed the name to R9.4 to see if the name was the issue.

#model_name     r9.10.1_400bps.nucleotide.9mer.template.model
#kit    r10.4.1_40bps

Is there a way to test a new R10 9-mer model in nanopolish?

I manage to do an event alignment using f5c eventalign using this model and most reads successfully aligned. As I am not sure of the best way to check the accuracy of signal alignments, I thought perhaps using the nanopolish variant caller could be a solution. Your thoughts are welcome.