search

nanoporetech / bonito

A PyTorch Basecaller for Oxford Nanopore Reads
https://nanoporetech.com/
Other
382 stars 118 forks source link

how to calibrate CTC-CRF model base qualities? #375

Open lpryszcz opened 7 months ago

lpryszcz commented 7 months ago

Hi, how to calibrate bonito trained models so base qualities correspond to expected error rate? For example config.toml for RNA004 sup models uses: 

[qscore]
scale = 0.9
bias = -0.1

How were those values obtained?

lpryszcz commented 3 months ago

any update on this? we're in the review process of basecalling model trained with bonito and we are unable to calibrate qscores therefore qscores from CTC-CRF models are not comparable with older models (flip-flop). I have opened similar issue in dorado repo