Question about tuning pretrained models

[najohink]

I would like to fine tune pretrained models to detect uridine bases. Would following the same pipeline of "basecall then train" work? I was thinking I would basecall sequences with Us and provide the corresponding reference.mmi with Us in it, then do the training. I am not sure if this will work, but ultimately I just wanted to be able to detect Us in my basecalled sequences.

-Does my dataset containing Us need to be perfect? If I have included sequences that don't contain Us, will it matter?

-Can I fine tune a model more than once? I noticed in an older discussion post https://github.com/nanoporetech/bonito/discussions/113 that someone wanted to train with multiple datasets, each with their own reference sequence. They were told to basecall each individually, and then merge the .npy files before proceeding to the train step. Would this be the same advice if I wanted to fine tune pretrained models with WT and KO datasets (one with and one without Us in their reference sequence)?

Thank you!

best,
S