search

theosanderson / mov_streamlit

1 stars 0 forks source link

Explain how to interpret results in the app #1

Open corneliusroemer opened 3 weeks ago

corneliusroemer commented 3 weeks ago

I'm not quite sure how to interpret the results. It would be great to have a little bit of context included on the app itself.

e.g. what does "Molnupiravir likeness" mean? A mutation can either be molnupiravir-like or not, but what do the log likelihoods mean in that context? What's a "neutral" result? Log likelihood of 0?

theosanderson commented 3 weeks ago

The numbers shown here only refer to evidence from the contexts in which the mutations occur. Molnupiravir evidence often comes from the basic classes of mutations (G>A, C>T) that occur on a branch. This calculator aims to provide additional evidence to confirm or point against a molnupiravir origin from the contexts in which those mutations occur, independent of the class of mutation. Therefore for each of the transitional classes of mutation it calculates the bayes factor for observing the contexts of the mutations under a model in which the mutations occur due to MOV and one in which the mutations occur due to typical evolution. A summed value of 0 would indicate those to are equally likely, so you should stick with your prior. A highly positive value points towards MOV from the contexts.

corneliusroemer commented 3 weeks ago

Thanks! Is there a reason you don't also include the mutation class in the calculation? A one stop shop would be very beneficial. (Should make a separate issue for that ;) )

theosanderson commented 3 weeks ago

This calculator is typically used by people who have branches that they already think are due to molnupiravir on the basis of their mutation classes, and the aim is simply to use contexts to confirm that. Factoring in the classes would create a risk of the prosecutor's fallacy. That said I have plans for a more comprehensive mutation spectrum analysis tool that would allow both to be toggled on and off.

corneliusroemer commented 3 weeks ago

So I'm not a typical user :p

I just plug in long branches

Do you have a CLI for the calculator? Would be nice to run on a large number of mutation sets, e.g. private mutations as output by Nextclade.