Running AlphaFold to Predict Protein Complexes from ChimeraX

[@danclayton5996]

Thanks for all the ChimeraX / Alphafold tutorials! Really useful, keep them coming! Docking the predicted structure with ligand binding would be great.

[@ashenafikiroswubshet5342]

Thanks those who built such an amazing program and definitely much helpful for every researchers globally. The very good thing about this software is is no barrier to use it. Non discriminatory at all. Much thankful.

[@haiyangcui2660]

very good introduction!!!

[@fedijebri394]

Loved the video

[@jamesflaumenhaft8223]

Great video, excited to get get started with ChimeraX! I am wondering if you of any way to perform the Global Distance Test (the metric used in CASP14). This would be incredibly helpful to my current research and would be interesting to be able to quantify the validity of these already image structures.

[@ucsfchimerax8387]

ChimeraX does not have the Global Distance Test metric from CASP14. But it would probably be easy for you to write some Python code to do it. Here are some ChimeraX Python code examples rbvi.github.io/chimerax-recipes/

[@caem747]

Thanks for the helpful video. Is it possible to let AlphaFold predict interactions of two proteins where I already have the .pdb file? Especially if my two proteins would exceed the 1000 residue GPU limit, could this help to circumvent the prediction limitation?

[@ryhisner]

This is a question I very much would like to know the answer to as well.

[@tomasfernandes7944]

Thank you so much! Has the heat map feature of AlphaFold been added? If so, what is the command? Thank you again! :)

[@SurfaceColor]

No. I plan to add AlphaFold heatmap display (residue-residue distance error plot) to ChimeraX. But there are hundreds of feature requests competing for time, and currently most of my time is going toward grant proposals to fund these developments. Maybe heatmap display in 6 months.

[@jinhanglin]

Thank you so much！

[@phantomcreamer]

I am guessing that confidence is because the protein you are looking at was published in pdb before alphafold2 was trained. I watched a presentation from the alfafold team and they said that would give a false level of confidence just like what you are showing. Those alpha helices and trimeric alignments should not be that high.

[@rodrigomaillard150]

This is really great. For homomeric proteins, i,.e., a homodimer, do you copy the amino acid sequence twice separated by a coma?

[@SurfaceColor]

Yes

[@sanjaisrao484]

Thank you

[@ryhisner]

I tried running a protein complex (TOM70 + HSP90alpha x2 + Cullin 5 + SARS-CoV-2 ORF9b) that researchers in a published study ran on Alphafold2. But the results I got look absolutely nothing like their results. Is this a a result of using ChimeraX? I only know how to use Alphafold using ChimeraX. Does anyone know of a good guide for using Alphafold on protein complexes without using ChimeraX?

[@swurlnplot]

Outstanding video. One question - if I close my computer // put it to sleep, will the computation continue?

[@ucsfchimerax8387]

If your computer sleeps Google Colab may disconnect halting your AlphaFold prediction. Unfortunately Google Colab does not document how long the connection has to remain idle before they disconnect. This is a big drawback of Colab. But since it is a free service or very cheap paid service (Colab Pro $10/month), it is hard to expect more. Google Colab is intended for interactive use.

[@matheushenriquereis1876]

Thanks for the tutorial! I just have one question. If I have to predict an eukaryote protein complex and don't check prokaryote button, the Alphafold multimer will run fine?

[@ucsfchimerax8387]

The latest AlphaFold 2.2.0 eliminated the "prokaryote" setting. The AlphaFold developers said it did not improve structures on average. The current ChimeraX daily build uses AlphaFold 2.2.0 and so it no longer has the prokaryote option. If you are using an older ChimeraX that has the prokaryote option, it still uses the newer AlphaFold 2.2.0 and simply ignores the setting of that option. If you run AlphaFold 2.1 outside of ChimeraX the multimer predictions will work fine even if you set the prokaryote option incorrectly. It will just change how the sequence alignments are assembled, so the resulting structure might be slightly different.

[@uzaysezen7319]

Fantastic tutorial! I immediately installed a daily build and folded the small subunit of my favorite dimer forming protein. I have been trying to fold the entire dimer which together is 834 aa long but the session times out after 5 hours. Should I keep trying?

[@tomasfernandes7944]

Hi! From my experience, you will need a Google Colab Pro subscription in order to perform the prediction of the entire dimer. Regular Google Colab allows you to run calculations up to 600 amino acids, whereas Google Colab Pro has allowed me to predict the structure of proteins up to ~1300 amino acids.

[@uzaysezen7319]

@@tomasfernandes7944 Thanks for the 600 aa info. I will try the pro account for a few months and see what kinds of results I get.

[@jamesflaumenhaft8223]

I am curious about how alphafold2 multimers will change the landscape of docking? How could one go about using alphafold to predict binding and protein-protein interaction?

[@jamesflaumenhaft8223]

Oh I see the most recent video

[@ucsfchimerax8387]

A recent article describes trying protein-protein docking of all yeast proteins using AlphaFold: "Computed structures of core eukaryotic protein complexes", (www.science.org/doi/10.1126/science.abm4805). AlphaFold only handles the standard 20 amino acids so cannot dock anything else.

[@madhurik8389]

could you please share best reference paper for beginner. if you have any on mind

[@barirazahid7180]

When new video is coming up ?

[@alonsovilca7013]

Where did u get the experimental file from

[@tomasfernandes7944]

Hi! The option to predict protein complexes from ChimeraX is no longer available? I have the version from 08/12/2021 and when I insert two protein sequences separated by a comma, it does not ask me if those proteins are prokaryote and an error message pops up. Do I have to install anything? Thank you!

[@SurfaceColor]

You need a ChimeraX daily build. The ChimeraX 1.3 release (Dec 8, 2021) only includes monomer prediction -- its code was frozen a month earlier before AlphaFold Multimer even came out. I'll fix the video description to explain that 1.3 does not have it.

[@ahmadkezzo]

Hello, is there a limit number of amino acids (in total) to be run to predict the multimeric structure, for example I need to run a prediction to know how a certain protein may interact with itself, so it may form a dimer, trimer, or oligomer. it is about 300 amino acids, do I need to make a run for every x-mer interaction, like for dimer to inter the AA sequence twice and run Chimera, and so on?? Thank you

[@ucsfchimerax8387]

The total number of amino acids that can be predicted is about 1000 and is limited by the GPU memory. Google Colab offers old GPUs, K80 with 12 GB, and V100, T4 with 16 GB. Modern GPUs can have 48 GB (Nvidia A40) or even 80 GB (Nvidia A100) and predict up to about 4000 amino acids. Here are example runs www.rbvi.ucsf.edu/chimerax/data/alphafold-jan2022/afspeed.html

[@fabioturetti2654]

The Webpage for ChimeraX is not available now?

[@ucsfchimerax8387]

UCSF servers running the ChimeraX site are down for maintenance Feb 28 - March 2. May be up sometime during these 3 days. Sorry.

[@adcoxh1]

What if one of your proteins is prokaryotic and the other is eukaryotic?

[@ucsfchimerax8387]

The "prokaryote" flag was removed by the AlphaFold developers around January 2022. If you use a current ChimeraX 1.4 version it will not show that option. If you want to use your current old ChimeraX just leave the prokaryote flag off -- that is how current AlphaFold versions -- they found that enabling the option did not improve predictions.

[@taslimanasrin4063]

What's the minimum RAM or set up required to run alphafold?

[@ucsfchimerax8387]

ChimeraX AlphaFold prediction runs on Google Colab servers so the memory on your computer is not used. In general I would recommend at least 8 GB of memory for general ChimeraX use.

[@taslimanasrin4063]

Thank you so much 🥰

[@patricial1231]

Are we supposed to have Python 3 already installed?

[@ucsfchimerax8387]

ChimeraX runs the AlphaFold prediction on Google Colab and it installs everything that is needed on the Colab virtual machine (including Python). ChimeraX itself includes Python 3.9 so it also does not need Python installed.

[@patricial1231]

@@ucsfchimerax8387 I keep getting an error that there is no module named alphafold.notebooks What does this mean? Am I supposed to load something after I go into the colab.ipynb ??

[@ucsfchimerax8387]

@@patricial1231 If you press the Predict button on the ChimeraX AlphaFold pane that is all you should need to do. There was an error the past day because Google moved the location of a library called jaxlib that AlphaFold needs. I fixed that error about 5 hours ago. The fix does not require a new ChimeraX. If it still does not work you can use ChimeraX menu Help / Report a Bug... and paste in the error from the Colab window into the bug report description.

[@patricial1231]

@@ucsfchimerax8387 it worked now, thanks!