APA Predict Variant Effects

Help & Support
1

Hey, confused about why I’m sometimes getting the error that:

columns = [ —>

c for c in df_scores.columns if c not in [‘variant_id’, ‘scored_interval’]

]

df_scores[columns]

AttributeError: ‘NoneType’ object has no attribute ‘columns’

I thought so long as the window contains a PAS (which it should), I should at least have one column in my output?

Thanks!

2

Hi @kylee , welcome to the community!

Usually the reason you wouldn’t have any columns is if there’s no scores for that particular region (i.e. these lines here), but if you’re saying its a region that should have some scores, then that’s confusing…

Do you have an example variant you could share that yields this error?

Thanks!

3

Hi @tward , thanks! Yeah, the lines you referred me to are why I was confused.

I was trying it out using ACTB with the 1 MB window, inserting SV40 (a known PAS) prematurely and observing the impact on termination. The thought process was that surely within that window there should be some scores…maybe I’m misunderstanding how the scorer works?

4

Hi @kylee , sorry for the delayed response!

There a couple of reasons as to why you aren’t getting any signal.

  1. Strength of the PAS site: Termination requires signal outside the PAS hexamer. If you only inserted the SV40 hexamer without its supporting downstream architecture, the model might view it as a weak or non-functional site. Make sure you’ve used the full ~200bp SV40 PAS
  2. Long context window: If the original ACTB termination site is still present within that window, the model may be drawn to the native, stronger signal. You could try to remove the native site (ie swap instead of just interesting the SV40 PAS)
  3. Ensure you’re working with the correct strand: ACTB is on the reverse strand

I’m interested in your use case - if you can send a code snippet, or let me know the sequence of the insertion and the insertion site, I can take a deeper look