Evaluating AlphaFold protein-protein binding with ChimeraX
Вставка
- Опубліковано 8 чер 2022
- We show how to display AlphaFold predicted aligned error for residues at the binding interface between two proteins, encapsulin and ferritin, using ChimeraX. Requires a ChimeraX daily build from June 9, 2022 or newer, or ChimeraX 1.5, not in ChimeraX 1.4. More details www.rbvi.ucsf.edu/chimerax/da...
- Наука та технологія
Thank you so much for your active dev of ChimeraX
Thanks! I'm working on a bunch of AlphaFold improvements. Stay tuned!
Once I have the contacts between structure A and B, is there the possibility to get a list of those contacts (to have a better overview)?
I’m also wondering whether bind affinities, and the binding energies could be predicted and shown (as there in many docking platforms)
@ucsfchimerax8387 I am having the same questions. Tried to play around the software but could not find a way to calculate binding energy. Is there a way of doing this?
Hi! Great explanation, thanks for showing the error plot function. I have version 1.5 but when I type the command for contacts between the two chains it says 'No residues specified for alphafold contacts'. Could you help me, what am I doing wrong? My guess is that it doesn't recognize that there are two separate chains?
Another question: I am not really satisfied with the position in which my protein A is now docked to the other, because I know from literature as well as from experiments I did in the lab that the binding is at a different domain... So would you recommend running alphafold again with some restrictions? Or, if I want try the docking manually: how can I separate the two strands?
The command used in the demonstration is "alphafold contacts /A to /B distance 8" which finds contacts between chain A and chain B. Your data probably uses other chain names (e.g. C and D) and so if you don't modify the command it says no residues were specified. For your second question, if AlphaFold does not predict the complex correctly there are no options to make it give a different result. You could predict each protein separately then open them and move one relative to the other by hand with the "Move model" mouse mode. If you have further questions ask on the ChimeraX mailing list.
Thanks for sharing!
Great video- Is there any reason you selected 8 as distance between /A and /B (two interacting proteins)? what is the range that we can consider as confident distance? can we go up to 15 ?
Thanks for the tutorial, it is very helpful. I wonder if ver. 1.5 is released now. I can only find the 1.4 and daily build versions. Do I need to delete the current 1.4 ver. and install the daily build one, or are there any commands to update? Thanks in advance.
ChimeraX 1.5 will be released in December 2022 or January 2023. The ChimeraX daily builds on the download page are always the most recent, including all the newest features. The daily builds do have newer features than ChimeraX 1.4 and the daily build is the only way to get those new features (unless it is a plugin you got from the ChimeraX Toolshed which will notify you of updates and have a button to install them). You can have multiple ChimeraX versions installed at the same time, no need to delete old versions.
Is it possible to find the average PAE for a given site? This would be a nice way of summarizing the evaluation of protein-protein interaction prediction alongside the visual producer
The pLDDT coloring of AlphaFold structures gives a good sense of confidence per-residue. PAE gives residue to residue positioning confidence. If you averaged for each residue over all other residues, if any part of the structure is bad then all the averages would be low. So I don't think an average of PAE would be useful.
@@ucsfchimerax8387 sorry for the unclear question, I was wondering whether there was a way to quickly average the PAE for an entire binding site (so multiple residues). While it is a nice visual to see the coloring of the contacts to gauge the certainty, it would also be Interesting to get an singular quantitive metric of the binding accuracy of a given area.
@@jamesflaumenhaft8223 I see. There is no capability to show an average. If you select the binding site residues and use command "alphafold contacts sel to sel distance 10" it will show many lines colored by confidence that will give a similar though more detailed picture. Since AlphaFold knows nothing about ligands, the binding site geometry is not trustworthy in general. You might think it would show an apo-binding site configuration but often it shows something closer to liganded configuration, perhaps because it was trained on many PDB structures with ligands.
Could you please demonstrate the tools one could use in chimera to demonstrate the potential effects an Amino acid change would have on a protien structure.
You could run AlphaFold on the unmutated and mutated sequences and compare the resulting structures. The ChimeraX swapaa command can replace single amino acid in a structure and the minimize mouse mode could remove the steric clashes that creates. These methods just give a tiny sense of how the new residue fits with its neighbors which usually says little about the effect of the mutation. Beyond those ideas, a whole lifetime of effort could be spent figuring out the effects of a mutation, usually involving lots of experimental work.
@@ucsfchimerax8387 I have the same issue with Hethro Sims.
Can the distances be measured for a trimer or can you only label distances between two proteins at once? Also What does the PAE domain coloring mean? I know the color key for pLDDDT but not for PAE domains.
Hi. Thank you for this great tutorial. I'm newbie in protein research. I have been using Alphafold with Chimerax interface for a while. I wonder how to do docking of two different proteins that I downloaded from UniProt to see which part of those proteins interact each other? Is this tutorial also about docking? or just about protein-protein binding? Because I need to learn which part of two proteins interact each other. I'm looking forward to see answers for that. Thanks!
You can have alphafold predict a complex of 2 or more proteins by pasting their sequences into the ChimeraX AlphaFold sequence pane separated by commas. This video shows an example ua-cam.com/video/6lXeCPuTePs/v-deo.html
@@ucsfchimerax8387 Thank you for this quick reply!
@@ucsfchimerax8387 what if we cannot download both sequences at the same time? I have both proteins already downloaded separately, but my computer is not powerful enough to download both of them at the same time like your video. Is there another way to simulate and find the place where these two proteins interact?
@@alonsovilca7013 If AlphaFold fails predicting the structure for two sequences that is probably because it is using old Google Colab GPUs that have limited memory (12-16 GB) and can only predict a structure of about 1200 amino acids. If you want to predict a larger structure you will need to setup AlphaFold yourself which is a good bit of work, and you will need a Linux machine with a GPU with more memory.
I followed the tutorial to the letter, when I try to plot the contacts I get the message "Structure best_model.pdb #1 does not have PAE data opened" even though the PAE is right there. Any advice?
I love your work by the way
The only thing I can think of is that when the PAE data was opened it was associated with a different structure. That doesn't make much sense since this video only uses one structure. But if you opened a different structure, for example, while waiting for the AlphaFold prediction to finish, then maybe the error plot shown was for that one. Here's the place in the video where you open the PAE data ua-cam.com/video/TMcjEecFHaI/v-deo.html and you'll see it has a menu to choose the structure the PAE data belongs with. Probably this is not your problem. If you see the problem again you can use ChimeraX menu entry Help / Report a Bug... and that may help us figure it out.
Just figured out the problem causing the "Structure ... does not have PAE data opened." A change in ChimeraX made on July 8 broke assigning PAE data to the structure, so ChimeraX daily builds from July 9 to August 2 were broken. The August 3, 2022 daily build and newer will work.
Hi when trying this I get the following error "Structure best_model.pdb #1 does not have PAE data opened" . Do you know how I can solve this?
Got it to work I needed to open the Aligned error plot
@@Troklo did you open it via command line or via Alphafold error menu?
I am having the same error
@@dandeelyonn The July 9 - August 2 ChimeraX daily builds had a bug causing this error even after you opened the PAE data. It is fixed in the August 3, 2022 ChimeraX daily build and newer.
@@ucsfchimerax8387 Thanks!
Thank you for your tutorial. I recently faced a problem (or a bug) that I could not get the final results file and got an error (NotImplementedError: A UTF-8 locale is required. Got ANSI_X3.4-1968) reported in colab.
This error happens when you enable energy minimization in the Options section of the ChimeraX AlphaFold panel. It is a new bug introduced probably in Google Colab (they updated from Python 3.7 to 3.8 last month). The AlphaFold prediction completes successfully but then fails packaging up the files for download because all shell command on Google Colab are broken, apparently because OpenMM which does the energy minimization for AlphaFold has changed the text encoding. I will look into a way to work around this error. Here is a ChimeraX bug report describing progress: www.rbvi.ucsf.edu/trac/ChimeraX/ticket/8313
@@ucsfchimerax8387 Hahaha😅 Actually, I was the person who reported this case #8313
@@ucsfchimerax8387 Thank you
@@zhenwang323 Yes, and I appreciate it! We always see the bug reports submitted with the Report a Bug... button that appears in ChimeraX for some errors, or with menu Help / Report a Bug.... We don't always see all the UA-cam comments.
@@zhenwang323 Thanks for reporting the problem! I spent half a day yesterday on it but my solutions did not work completely so I am working on it again today. When I fix it no update of ChimeraX will be needed since the fix will be in a script that ChimeraX gets from GitHub each time a prediction is run. So hopefully predictions with energy minimization will start working today or tomorrow.
When I use the contacts command it doesn't colour the contacts by PAE, how can I change that?
The "alphafold contacts" command always colors the pseudobonds it displays between CA atoms. They might all appear to be the same color if the PAE values for all of them are nearly the same. There is another ChimeraX command "contacts" that finds close atoms but does not use PAE.
Excellent videos. But I don't understand PAE plot. It would be helpful if you said a little more about it. For example, choose some points in the plot and say exactly what they mean
The PAE values are defined by AlphaFold and described in their publications and is not too intuitive. I have a brief explanation here www.rbvi.ucsf.edu/chimerax/data/pae-apr2022/pae.html. Unfortunately the videos reach more people if they are shorter and do not try to cover all the background material since typical viewers have a short attention span. So my aim in future videos is more along the lines of reducing the content so it can benefit more researchers rather than expanding it.