Colab pro + failing to predict >2000 residues

[HBCH1]

Hi, I am using AlphaFold 2 (Colab pro +) to predict some proteins with long sequences. While it successfully predicted the structures of shorter sequences, it keeps failing with longer ones even tho I upgraded to colab pro + to have higher memory.
Here is what I am getting:

Is there a way to fix this issue or is AlphaFold 2 not currently able to predict >2000 bp sequences?
Thanks in advance!

[Htomlinson14]

Hi @ HBCH1 what GPU were you using in this session?

[Augustin-Zidek]

We recently updated AlphaFold Colab notebook (see a0b0cd9) to handle longer sequences better. Could you try with the updated code?

[HBCH1]

Hi @ HBCH1 what GPU were you using in this session?

I was using Premium GPU with Colab Pro plus @Htomlinson14

[HBCH1]

Hi @Augustin-Zidek , thanks for the link - I was not able to upload the page associated with the link you provided. Where can I find the updated notebook?

[Augustin-Zidek]

Sorry, I fixed the link in my previous comment. The notebook was updated in the repository, so the standard Colab link points to the updated version

[Augustin-Zidek]

Closing this issue now, feel free to re-open if there is anything we could help with.

[HBCH1]

@Augustin-Zidek I have tried running my 2750 bp sequences on the new version but I had two issues:

1. msa was taking a very long time to search the data bases (~35 hours)
2. the prediction failed

Do you mind advising on how to fix those issues?