Beginner Help: Issues with "Predict Structures (GPU stage)" and Running AlphaPulldown on Google Colab

[Yuan-323]

Hi everyone,

I'm a beginner learning how to use AlphaPulldown by following the guide on the official website. During the process, I encountered two issues that I hope someone can help me resolve:

About Predict Structures (GPU stage)
When I tried running the step Predict structures (GPU stage), the guide mentioned specifying --data_dir=.
I understand that this step requires the AlphaFold databases, but these databases are extremely large (over a terabyte). Since I am running AlphaPulldown on my local computer and don’t have the storage capacity to download the entire databases, my learning progress is currently stuck at this step.
I’d like to ask if there are alternative ways to complete this step, such as using partial databases or accessing the databases directly over the internet without downloading everything locally?

About Running AlphaPulldown on Google Colab
Considering the limitations of my local setup, I was wondering if it is possible to run AlphaPulldown on Google Colab?
Colab provides free GPU resources and would make it easier to collaborate. If AlphaPulldown can be used on Colab, are there any detailed tutorials or steps available to help set this up?

I’d greatly appreciate any advice or guidance on these questions! Thank you so much in advance for your time and support! 😊

[jkosinski]

Hello,
To avoid installing large sequence databases, you can:

1. use precalculated features from our database https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#features-database
   or
2. use MMSeqs2 mode which will use a ColabFold/MMSeqs2 server to calculate MSAs: https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#13-run-using-mmseqs2-and-colabfold-databases-faster

It should be technically possible to run AlphaPulldown on Google Colab but we don't provide it from our side yet.

[Yuan-323]

Hello, Thank you for your reply. I would appreciate it if you could kindly clarify the following for me: I have already completed the first step of using precalculated features from the database and downloaded several protein files. However, I am a bit confused about the second step when running predict structure. It seems that the command always requires access to AlphaFold databases using the --data_dir=<path to alphafold databases> argument. Since downloading these files locally takes up a significant amount of space, I am wondering if there's an alternative solution. I noticed that the option to use remotely-run MMseqs2 and ColabFold databases is mentioned only in Step 1: Compute multiple sequence alignment (MSA) and template features (CPU stage). Could you please advise on how to handle this issue? Thank you for your time and support! Best regards, Yuan Yuan Liu PhD Student University College London Division of Medicine The Rayne Building 5 University Street London WC1E 6JF Major: Drug Design Student Number: 22038233 Pronouns He/Him Email: ***@***.*** Tel: +44 7536143785

…

________________________________ From: Jan Kosinski ***@***.***> Sent: 06 January 2025 20:49 To: KosinskiLab/AlphaPulldown ***@***.***> Cc: Liu, Yuan ***@***.***>; Manual ***@***.***> Subject: Re: [KosinskiLab/AlphaPulldown] Beginner Help: Issues with "Predict Structures (GPU stage)" and Running AlphaPulldown on Google Colab (Discussion #473) ⚠ Caution: External sender Hello, To avoid installing large sequence databases, you can: 1. use precalculated features from our database https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#features-database or 2. use MMSeqs2 mode which will use a ColabFold/MMSeqs2 server to calculate MSAs: https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#13-run-using-mmseqs2-and-colabfold-databases-faster It should be technically possible to run AlphaPulldown on Google Colab but we don't provide it from our side yet. — Reply to this email directly, view it on GitHub<#473 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BNDBGTIWJNMYN3XZ65DLKRT2JLT6NAVCNFSM6AAAAABUPRZUGKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCNZVGM3TENA>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[jkosinski]

Hello,

Indeed, that was not fully explained. I updated the manual now:

• in https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#script-execution-structure-prediction:
  Instead of provide a path to the directory containing params directory with AlphaFold paramaters (see 0. Alphafold-databases of the installation part). Sequence and PDB databases are not needed at this stage.
• in https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#alphafold-databases:
  Note
  Since the local installation of all genetic databases is space-consuming, you can alternatively use the remotely-run MMseqs2 and ColabFold databases. Follow the corresponding instructions. However, for AlphaPulldown to function, you must download the parameters stored in the params/ directory of the AlphaFold database by downloading and executing this script: https://github.com/google-deepmind/alphafold/blob/main/scripts/download_alphafold_params.sh

[Yuan-323]

Hi， Thank you so much for your detailed response and for updating the manual based on my question. I truly appreciate the effort you put into clarifying the steps and providing additional resources. Recently, I have been trying to install and run AlphaPulldown on Colab. https://colab.research.google.com/drive/1uOdIVEgifv7VTLYv8qqXtopE2RqhP8eZ?usp=sharing This is my progress so far. As a beginner in programming, I am aware that there might be many shortcomings in my code. If you have the time to take a look or leave any suggestions, I would be very grateful! Your feedback would mean a lot to me as I continue learning and improving. Thank you once again for your support and dedication to helping users like me.😁 Best regards, Yuan Yuan Liu PhD Student University College London Division of Medicine The Rayne Building 5 University Street London WC1E 6JF Major: Drug Design Student Number: 22038233 Pronouns He/Him Email: ***@***.*** Tel: +44 7536143785

…

________________________________ From: Jan Kosinski ***@***.***> Sent: 08 January 2025 8:00 To: KosinskiLab/AlphaPulldown ***@***.***> Cc: Liu, Yuan ***@***.***>; Manual ***@***.***> Subject: Re: [KosinskiLab/AlphaPulldown] Beginner Help: Issues with "Predict Structures (GPU stage)" and Running AlphaPulldown on Google Colab (Discussion #473) ⚠ Caution: External sender Hello, Indeed, that was not fully explained. I updated the manual now: * in https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#script-execution-structure-prediction: Instead of provide a path to the directory containing params directory with AlphaFold paramaters (see 0. Alphafold-databases<https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#installation> of the installation part). Sequence and PDB databases are not needed at this stage. * in https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#alphafold-databases: Note Since the local installation of all genetic databases is space-consuming, you can alternatively use the remotely-run MMseqs2 and ColabFold databases<https://github.com/sokrypton/ColabFold>. Follow the corresponding instructions<https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#13-run-using-mmseqs2-and-colabfold-databases-faster>. However, for AlphaPulldown to function, you must download the parameters stored in the params/ directory of the AlphaFold database by downloading and executing this script: https://github.com/google-deepmind/alphafold/blob/main/scripts/download_alphafold_params.sh — Reply to this email directly, view it on GitHub<#473 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BNDBGTJFTKLSVYJENJYMSE32JTLLTAVCNFSM6AAAAABUPRZUGKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCNZXGAYDANY>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>