Hi Anthony,

  Predicting a dimer where the monomer is 1182 residues, so a total of 2364 residues is guaranteed to run out of memory on Colab or Colab Pro -- the GPUs they offer don't have enough memory (at most 16 GB).  The web page on AlphaFold performance test cases shows that only an Nvidia A40 with 48 Gbytes of memory handles over 2100 residues.  That web page gives results for running AlphaFold via ChimeraX on Colab, but also running AlphaFold installed on our machines at UCSF with either an A40 GPU or an RTX 3090 GPU (24 GB memory).  The "no templates" means structure templates were not used when running on our UCSF machines.

Tom


On Jul 24, 2022, at 1:35 AM, Anthony Morgan <anthony.morgan@pharm.ox.ac.uk> wrote:

Dear Tom,

Thanks so very much, this is low-hanging fruit for you, but invaluable for me as I am a newbie.  Really appreciate it. The Test link is really illuminating too thank you.

  • I had wanted to try a dimer of 1182 residues each, but I am doubting that this is going to work, even with the paid service. Agreed?
  • On the test page, one of the runs is annotated as “no templates”. I cannot see how you can add templates to the Alphafold panel. I only see Energy Minimization under Options. Am I missing something?

I’ve now got the pdb of the one model, thank you. It’s obvious once it’s pointed out, but I couldn’t see the wood for the trees.

Really appreciate your time.

Kind regards,

Anthony


On 24 Jul 2022, at 00:37, Tom Goddard <goddard@sonic.net> wrote:

Hi Anthony,

  AlphaFold predictions on Google Colab of 1200 amino acids often fail as you can see from test cases on this web page

https://www.rbvi.ucsf.edu/chimerax/data/alphafold-jan2022/afspeed.html

With free Google Colab you often get a very old Nvidia K80 GPU with 12 Gbytes of memory.  With Colab Pro ($10/month) I usually get an Nvidia P100 GPU with 16 Gbytes of memory which can handle slightly bigger proteins.  You can check what GPU you got by pressing the "+ Code" button on the Google Colab window, then entering the command "!nvidia-smi" and pressing the play button, which gives output like this

Sat Jul 23 23:28:55 2022 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |===============================+======================+======================| | 0 Tesla P100-PCIE... Off | 00000000:00:04.0 Off | 0 | | N/A 38C P0 34W / 250W | 16259MiB / 16280MiB | 0% Default | | | | N/A | +-------------------------------+----------------------+----------------------+
  To download a completed model after an AlphaFold error (typically out of memory), click on Files in the Google Colab window, then click on the predicted structure, "af1182_unrelaxed_model_1.pdb" in your image, and click the "..." to the right of the file and choose Download from the menu that appears.  The file will then appear on your local computer in 

~/Downloads/ChimeraX/AlphaFold/prediction_1

  I'll see if I can add to the ChimeraX ColabFold output the GPU and also make it automatically download partial results in case an error occurs.

Tom


On Jul 23, 2022, at 4:41 AM, Anthony Morgan via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:

Dear All,

I’m new to Chimera X and to Colab Alphafold (AF), hope you can help please.

  • I’m running Chimera X 1.4 on a Macbook Air.
  • I want the predicted structure of a fairly large protein (1182 residues).
  • When I first ran Colabfold_predict, it terminated after 90 min (RAM issue).
  • I re-ran the same sequence *without * Energy Minimization.
  • After 20 mins, it successfully returned the first model (see attached), but ran into RAM issues again.

Two questions:

  • There is no ‘Prediction’ folder in my Downloads, so how can I download the *.pdb file of model #1 that it clearly has generated?
  • If I cannot do this, will an upgrade to Colab Pro solve this issue for this size of protein?

Thanks for advice.

Kind regards,

Anthony

<Screen Shot 2022-07-23 at 12.17.14.png>




_______________________________________________
ChimeraX-users mailing list
ChimeraX-users@cgl.ucsf.edu
Manage subscription:
https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users