search

nanoporetech / taiyaki

Training models for basecalling Oxford Nanopore reads
https://nanoporetech.com/
Other
115 stars 42 forks source link

Evaluation Method #91

Closed techsavy12 closed 4 years ago

techsavy12 commented 4 years ago

Hello, I am able to obtain basecall.hdf5 files from my newly trained model. I believe that this file provides the modified probability of each base of each reads. However, I was wondering if there are any method such as ROC or AUC curves available to evaluate my model.

techsavy12 commented 4 years ago

I have been looking through megalodon but I'm unsure as to how we would input the output obtained from taiyaki into megalodon. Are there other evaluation methods available. Your help is very much appreciated. Thank You!

marcus1487 commented 4 years ago

See relevant megalodon argument docs here (https://nanoporetech.github.io/megalodon/advanced_arguments.html#guppy-backend-argument) and here (https://nanoporetech.github.io/megalodon/advanced_arguments.html#taiyaki-backend-arguments). Also see the Taiyaki dump_json command to save a model to use with guppy (can use via megalodon with this output as well). Megalodon provides the highest accuracy modified base calls and is thus the recommended method.