search

jbloomlab / polyclonal

Model mutational escape from polyclonal antibodies.
Other
3 stars 0 forks source link

Specifying epitopes #25

Closed timcyu closed 2 years ago

timcyu commented 2 years ago

@jbloom I've attached a notebook for your review that experiments with the incorrect epitope scenarios described in #9, and provides some guidelines for how to specify the number of epitopes in the "final" model. Let me know if you have any thoughts!

timcyu commented 2 years ago

@jbloom I've addressed your comments. Let me know if there's anything that's unclear or if you have more thoughts!

timcyu commented 2 years ago

@jbloom here is the newly formatted specify_epitopes notebook. I've separated the effects of varying number of epitopes and seeding. I've also focused on showing correlation to the true epitope beta values and how good IC90 prediction is. I think I can improve the writing more, but let me know what you think of this format.

timcyu commented 2 years ago

@jbloom Thanks for the comments. I've made the fitting with manual seeding more concise and added the correlation scatter plots for better comparison.

I also re-fit all of the models in this notebook as I noticed the results are a bit different from what I had fit before the holiday (probably due to change in the site spread regularization calculation). Most differences are barely noticeable, although the manual seeding for the 2-epitope case looks slightly worse. Regardless, the overall message isn't affected. I'll just keep in mind that once the code is finalized, I should update the models in all the previous notebooks.