Thank you so much for posting this. It was really user friendly and easy-to-follow. It also explained some of the "insufficient GPU" errors I was getting from the AlphaFold2 colab notebook. It is important (as you noted) to install the "Daily Build" "chimerax-daily.exe" file as AlphaFold wasn't added before the 1.2.5 release. Thank you!
Good video. But I have around 100 different protein sequences, so how to automate this prediction? Can we paste all the 100 sequences in one single prediction by a separator? Can you suggest me a solution to do for multiple sequences?
Thank you. Please clear that the newly predicted structure in AlphaFold database needs to be validated through Ramachandra plot or needs to be refinement or not? Or can I directly use them for further docking procedures without refinement!?? (I've seen the models after refinement clash score, molprobity is risen instead of lowered) Plz clarify me.
I have two peptides with 17 residues each, one is head to tail cyclic peptide and the other is stapled peptide. Does alphafold2 work with chemically modified peptides?
Hi, thank you for sharing this video. I am wondering if you plan to release v1.3 sometime soon. I think currently v1.2.5 is the available version of ChimeraX that doesn't utilize AlphaFold.
Use a ChimeraX daily build from the download page if you want to try the AlphaFold tool. ChimeraX 1.3 will be released probably at end of October and will also include it.
The video uses a sequence affiliated with a pre-existing PDB... anyway to use the same workflow with a protein sequence that does not have a PDB entry?
Yes you can paste in a sequence in the ChimeraX AlphaFold panel. Select "paste" from the Sequence menu in the AlphaFold panel (see time 1:06 in this video) and an entry box for the sequence will display. I will make another video showing how to predict a structure starting with just a sequence and will post the link to the new video in reply to this question when it is available.
Here is a video showing how to fit an AlphaFold database structure into a cryoEM map without using a preexisting PDB model. th-cam.com/video/Asp2Sg-AN-g/w-d-xo.html
I have a question I'm trying to predict two proteins with the linker between but the alpha fold has a disk issue. I don't understand is about my disk because I have 1 tb space to use but when it makes model number 3 it start stopping. can you please describe this?
ChimeraX AlphaFold prediction runs on a Google Colab server and the error you see is because that Colab server is out of disk space. The initial disk space of these free virtual machines is around 80 Gbytes and I have seen it run out of space when the sequence has hundreds of thousands of sequence alignment matches. With Colab Pro paid service from Google that costs $10 per month the disk space is about 160 Gbytes and I have not seen it run out of disk space, but I still see it fail for all long sequences > 800 amino acids because the GPU memory is not adequate.
To run longer sequences than 800 amino acids you would need to run AlphaFold on more powerful hardware than what ChimeraX uses from the free Google Colab servers. I have run it on a 1600 amino acid sequence on our UC San Francisco cluster with Nvidia A40 GPU which has 48 Gbytes of video memory and it took 20 hours to complete (myomesin, uniprot MYOM1_HUMAN). That is a $10,000 GPU, running on a $25000 compute node, so it is not equipment that everyone has. This is a significant limitation of AlphaFold.
@@ucsfchimerax8387 Hello, I am running a sequence with 257 aa, would the free colab work or would I need ColabPro? When I was attempting to running the sequence, I ran into the CUDA error you mentioned. Please let me know what you think, thank you!
@@xitlallizavala1422 257 amino acids will run fine on free colab. Any GPU can run that. If it is producing an error it is very likely not out of memory. You can use ChimeraX menu Help / Report a Bug... and copy and paste the error that AlphaFold gives. Or you can just run AlphaFold outside ChimeraX -- I'd suggest trying the ColabFold server which runs the predictions much faster.
No. The Google Colab version of AlphaFold that ChimeraX runs does not use any structure templates and uses a reduced set of sequence databases compared to the AlphaFold used for CASP14 and for making the EBI AlphaFold database structures. Take a look here for the limitations: colab.research.google.com/github/deepmind/alphafold/blob/main/notebooks/AlphaFold.ipynb
@@ucsfchimerax8387 thanks! Do you happen to know what templates are used for the database that they published? Seems like chimera also can fetch those.
@@mashakarelina5110 I believe they used the entire PDB database snapshot in computing the structures in the AlphaFold database. Details are in the AlphaFold database journal article.
![CAMPปลิ้น | EP.83[2/2] ประสบการณ์ขนหัวลุกจากตำนานวงการผี!](http://i.ytimg.com/vi/T_kFecQw7Ds/mqdefault.jpg)
Thank you so much for posting this. It was really user friendly and easy-to-follow. It also explained some of the "insufficient GPU" errors I was getting from the AlphaFold2 colab notebook. It is important (as you noted) to install the "Daily Build" "chimerax-daily.exe" file as AlphaFold wasn't added before the 1.2.5 release. Thank you!
Good video. But I have around 100 different protein sequences, so how to automate this prediction? Can we paste all the 100 sequences in one single prediction by a separator? Can you suggest me a solution to do for multiple sequences?
I've installed chimeraX (macos high sierra) but not show structure prediction
Great video! Very informative.
🙏 thank you for this!
Thank you. Please clear that the newly predicted structure in AlphaFold database needs to be validated through Ramachandra plot or needs to be refinement or not? Or can I directly use them for further docking procedures without refinement!?? (I've seen the models after refinement clash score, molprobity is risen instead of lowered) Plz clarify me.
How can I save it as a pdb file. I can only save it as py and ipynb, which I cannot open in chimeraX
I have two peptides with 17 residues each, one is head to tail cyclic peptide and the other is stapled peptide. Does alphafold2 work with chemically modified peptides?
AlphaFold only works on standard amino acids. This is a limitation of AlphaFold, not specific to ChimeraX.
Hi, thank you for sharing this video. I am wondering if you plan to release v1.3 sometime soon. I think currently v1.2.5 is the available version of ChimeraX that doesn't utilize AlphaFold.
Use a ChimeraX daily build from the download page if you want to try the AlphaFold tool. ChimeraX 1.3 will be released probably at end of October and will also include it.
@@ucsfchimerax8387 Thank you. I appreciate your efforts.
The video uses a sequence affiliated with a pre-existing PDB... anyway to use the same workflow with a protein sequence that does not have a PDB entry?
Yes you can paste in a sequence in the ChimeraX AlphaFold panel. Select "paste" from the Sequence menu in the AlphaFold panel (see time 1:06 in this video) and an entry box for the sequence will display. I will make another video showing how to predict a structure starting with just a sequence and will post the link to the new video in reply to this question when it is available.
Here is a video showing how to fit an AlphaFold database structure into a cryoEM map without using a preexisting PDB model. th-cam.com/video/Asp2Sg-AN-g/w-d-xo.html
Here is one more video showing how to run AlphaFold from just a sequence for building into a cryoEM map. th-cam.com/video/x-9urTQ1rJY/w-d-xo.html
I have a question I'm trying to predict two proteins with the linker between but the alpha fold has a disk issue. I don't understand is about my disk because I have 1 tb space to use but when it makes model number 3 it start stopping. can you please describe this?
ChimeraX AlphaFold prediction runs on a Google Colab server and the error you see is because that Colab server is out of disk space. The initial disk space of these free virtual machines is around 80 Gbytes and I have seen it run out of space when the sequence has hundreds of thousands of sequence alignment matches. With Colab Pro paid service from Google that costs $10 per month the disk space is about 160 Gbytes and I have not seen it run out of disk space, but I still see it fail for all long sequences > 800 amino acids because the GPU memory is not adequate.
To run longer sequences than 800 amino acids you would need to run AlphaFold on more powerful hardware than what ChimeraX uses from the free Google Colab servers. I have run it on a 1600 amino acid sequence on our UC San Francisco cluster with Nvidia A40 GPU which has 48 Gbytes of video memory and it took 20 hours to complete (myomesin, uniprot MYOM1_HUMAN). That is a $10,000 GPU, running on a $25000 compute node, so it is not equipment that everyone has. This is a significant limitation of AlphaFold.
@@ucsfchimerax8387 Hello, I am running a sequence with 257 aa, would the free colab work or would I need ColabPro? When I was attempting to running the sequence, I ran into the CUDA error you mentioned. Please let me know what you think, thank you!
@@xitlallizavala1422 257 amino acids will run fine on free colab. Any GPU can run that. If it is producing an error it is very likely not out of memory. You can use ChimeraX menu Help / Report a Bug... and copy and paste the error that AlphaFold gives. Or you can just run AlphaFold outside ChimeraX -- I'd suggest trying the ColabFold server which runs the predictions much faster.
Hi, just wondering - when a structure exists in PDB - does that structure not get used as a template for structure prediction?
No. The Google Colab version of AlphaFold that ChimeraX runs does not use any structure templates and uses a reduced set of sequence databases compared to the AlphaFold used for CASP14 and for making the EBI AlphaFold database structures. Take a look here for the limitations: colab.research.google.com/github/deepmind/alphafold/blob/main/notebooks/AlphaFold.ipynb
@@ucsfchimerax8387 thanks! Do you happen to know what templates are used for the database that they published? Seems like chimera also can fetch those.
@@mashakarelina5110 I believe they used the entire PDB database snapshot in computing the structures in the AlphaFold database. Details are in the AlphaFold database journal article.