Hi ChimeraX Dev
I wondering if I can run oligomer prediction on ChimeraX AlphaFold features. As we can do it with others colabfold version out there. Thank you
Not sure why this text does not show up in the text of the message but only in the summary shown by Mail.app, but apparently the text of the message should be:
On Nov 14, 2021, at 9:48 PM, Ichcharisma via ChimeraX-users chimerax-users@cgl.ucsf.edu wrote:
Hi ChimeraX Dev
I wondering if I can run oligomer prediction on ChimeraX AlphaFold features. As we can do it with others colabfold version out there. Thank you _______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
ChimeraX does not yet allow running AlphaFold-Multimer the version of AlphaFold released November 2, 2021 that can predict structures for molecular assemblies. We want to offer this, although a ChimeraX version will only be a slight convenience over running an AlphaFold-Multimer Google Colab notebook with a web browser without ChimeraX.
The ChimeraX Google Colab AlphaFold predictions are limited in how long a sequence they can handle to about 800-1000 amino acids. This is because the old Google Colab GPUs (Nvidia K80 and T4) have only 16 Gbytes of memory and AlphaFold runs out of memory. When you predict a molecular assembly with AlphaFold-Multimer, the sum of all sequence lengths will have the same limit. So only small assemblies will be possible. To predict longer sequences or larger assemblies you need a GPU with more memory such as an Nvidia A100 with 80 Gbytes, or an A40 with 48 Gbytes. I don't know if Google Colab offers or will offer those in the future. I am currently working on running AlphaFold-Multimer on the UCSF cluster with A40 GPUs to see how well it works and what sequence length it can handle. The memory use is said to increase as the square of the sequence length. We are looking at whether ChimeraX could run AlphaFold predictions on the UCSF Wynton cluster but we probably do not have the budget to pay for that since it would probably cost about $100,000 per year to maintain about 16 high-end GPUs to meet the demand of ChimeraX users. So we are still figuring out what AlphaFold services ChimeraX can offer.
Tom
On Nov 15, 2021, at 9:13 AM, Eric Pettersen via ChimeraX-users chimerax-users@cgl.ucsf.edu wrote:
Not sure why this text does not show up in the text of the message but only in the summary shown by Mail.app, but apparently the text of the message should be:
On Nov 14, 2021, at 9:48 PM, Ichcharisma via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote:
Hi ChimeraX Dev
I wondering if I can run oligomer prediction on ChimeraX AlphaFold features. As we can do it with others colabfold version out there. Thank you _______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
(The original text was with white front color. It shows up for me once the texts are selected... )
Come in to just say that the alphafold2 multimer is still under active debugging. We had faced issue #227 https://github.com/deepmind/alphafold/issues/227 and it is solved days ago. But we are now facing issue #236 https://github.com/deepmind/alphafold/issues/236 frequently, mostly with multimers whose total length exceeds 1000 amino acids. From the discussion, these clashes are also seen in the Colab version. The developer says "We think we may have identified the issue with overlapping domains but need to test a bit more. It will take us a few days at least to be sure." With the current version, it will probably waste a lot of computing time. So one might want to hold the integration to ChimeraX until #236 is solved.
Best, Roden
On Mon, Nov 15, 2021 at 10:32 PM Tom Goddard via ChimeraX-users < chimerax-users@cgl.ucsf.edu> wrote:
ChimeraX does not yet allow running AlphaFold-Multimer the version of AlphaFold released November 2, 2021 that can predict structures for molecular assemblies. We want to offer this, although a ChimeraX version will only be a slight convenience over running an AlphaFold-Multimer Google Colab notebook with a web browser without ChimeraX.
The ChimeraX Google Colab AlphaFold predictions are limited in how long a sequence they can handle to about 800-1000 amino acids. This is because the old Google Colab GPUs (Nvidia K80 and T4) have only 16 Gbytes of memory and AlphaFold runs out of memory. When you predict a molecular assembly with AlphaFold-Multimer, the sum of all sequence lengths will have the same limit. So only small assemblies will be possible. To predict longer sequences or larger assemblies you need a GPU with more memory such as an Nvidia A100 with 80 Gbytes, or an A40 with 48 Gbytes. I don't know if Google Colab offers or will offer those in the future. I am currently working on running AlphaFold-Multimer on the UCSF cluster with A40 GPUs to see how well it works and what sequence length it can handle. The memory use is said to increase as the square of the sequence length. We are looking at whether ChimeraX could run AlphaFold predictions on the UCSF Wynton cluster but we probably do not have the budget to pay for that since it would probably cost about $100,000 per year to maintain about 16 high-end GPUs to meet the demand of ChimeraX users. So we are still figuring out what AlphaFold services ChimeraX can offer.
Tom
On Nov 15, 2021, at 9:13 AM, Eric Pettersen via ChimeraX-users < chimerax-users@cgl.ucsf.edu> wrote:
Not sure why this text does not show up in the text of the message but only in the summary shown by Mail.app, but apparently the text of the message should be:
On Nov 14, 2021, at 9:48 PM, Ichcharisma via ChimeraX-users < chimerax-users@cgl.ucsf.edu> wrote:
Hi ChimeraX Dev
I wondering if I can run oligomer prediction on ChimeraX AlphaFold features. As we can do it with others colabfold version out there. Thank you _______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription:
https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chime...
Thanks Roden for the tips on AlphaFold-Multimer problems. I just got the AlphaFold-Multimer databases setup on our UCSF cluster today and am about to run some test cases and your advise will help me avoid some problem cases.
I gave a talk on AlphaFold and ChimeraX for SBGrid last week that discusses some of the problems with AlphaFold
https://www.rbvi.ucsf.edu/chimerax/data/alphafold-nov2021/af_sbgrid.html
Tom
On Nov 15, 2021, at 12:14 PM, Roden Deng Luo deng.luo@kaust.edu.sa wrote:
(The original text was with white front color. It shows up for me once the texts are selected... )
Come in to just say that the alphafold2 multimer is still under active debugging. We had faced issue #227 https://github.com/deepmind/alphafold/issues/227 and it is solved days ago. But we are now facing issue #236 https://github.com/deepmind/alphafold/issues/236 frequently, mostly with multimers whose total length exceeds 1000 amino acids. From the discussion, these clashes are also seen in the Colab version. The developer says "We think we may have identified the issue with overlapping domains but need to test a bit more. It will take us a few days at least to be sure." With the current version, it will probably waste a lot of computing time. So one might want to hold the integration to ChimeraX until #236 is solved.
Best, Roden
On Mon, Nov 15, 2021 at 10:32 PM Tom Goddard via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote: ChimeraX does not yet allow running AlphaFold-Multimer the version of AlphaFold released November 2, 2021 that can predict structures for molecular assemblies. We want to offer this, although a ChimeraX version will only be a slight convenience over running an AlphaFold-Multimer Google Colab notebook with a web browser without ChimeraX.
The ChimeraX Google Colab AlphaFold predictions are limited in how long a sequence they can handle to about 800-1000 amino acids. This is because the old Google Colab GPUs (Nvidia K80 and T4) have only 16 Gbytes of memory and AlphaFold runs out of memory. When you predict a molecular assembly with AlphaFold-Multimer, the sum of all sequence lengths will have the same limit. So only small assemblies will be possible. To predict longer sequences or larger assemblies you need a GPU with more memory such as an Nvidia A100 with 80 Gbytes, or an A40 with 48 Gbytes. I don't know if Google Colab offers or will offer those in the future. I am currently working on running AlphaFold-Multimer on the UCSF cluster with A40 GPUs to see how well it works and what sequence length it can handle. The memory use is said to increase as the square of the sequence length. We are looking at whether ChimeraX could run AlphaFold predictions on the UCSF Wynton cluster but we probably do not have the budget to pay for that since it would probably cost about $100,000 per year to maintain about 16 high-end GPUs to meet the demand of ChimeraX users. So we are still figuring out what AlphaFold services ChimeraX can offer.
Tom
On Nov 15, 2021, at 9:13 AM, Eric Pettersen via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote:
Not sure why this text does not show up in the text of the message but only in the summary shown by Mail.app, but apparently the text of the message should be:
On Nov 14, 2021, at 9:48 PM, Ichcharisma via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote:
Hi ChimeraX Dev
I wondering if I can run oligomer prediction on ChimeraX AlphaFold features. As we can do it with others colabfold version out there. Thank you _______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chime... https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
This message did not make it into the ChimeraX mailing list archive, possibly because our mailing list server was down.
Thanks Roden for the tips on AlphaFold-Multimer problems. I just got the AlphaFold-Multimer databases setup on our UCSF cluster today and am about to run some test cases and your advise will help me avoid some problem cases.
I gave a talk on AlphaFold and ChimeraX for SBGrid last week that discusses some of the problems with AlphaFold
https://www.rbvi.ucsf.edu/chimerax/data/alphafold-nov2021/af_sbgrid.html https://www.rbvi.ucsf.edu/chimerax/data/alphafold-nov2021/af_sbgrid.html
Tom
On Nov 15, 2021, at 12:14 PM, Roden Deng Luo deng.luo@kaust.edu.sa wrote:
(The original text was with white front color. It shows up for me once the texts are selected... )
Come in to just say that the alphafold2 multimer is still under active debugging. We had faced issue #227 https://github.com/deepmind/alphafold/issues/227 and it is solved days ago. But we are now facing issue #236 https://github.com/deepmind/alphafold/issues/236 frequently, mostly with multimers whose total length exceeds 1000 amino acids. From the discussion, these clashes are also seen in the Colab version. The developer says "We think we may have identified the issue with overlapping domains but need to test a bit more. It will take us a few days at least to be sure." With the current version, it will probably waste a lot of computing time. So one might want to hold the integration to ChimeraX until #236 is solved.
Best, Roden
On Mon, Nov 15, 2021 at 10:32 PM Tom Goddard via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote: ChimeraX does not yet allow running AlphaFold-Multimer the version of AlphaFold released November 2, 2021 that can predict structures for molecular assemblies. We want to offer this, although a ChimeraX version will only be a slight convenience over running an AlphaFold-Multimer Google Colab notebook with a web browser without ChimeraX.
The ChimeraX Google Colab AlphaFold predictions are limited in how long a sequence they can handle to about 800-1000 amino acids. This is because the old Google Colab GPUs (Nvidia K80 and T4) have only 16 Gbytes of memory and AlphaFold runs out of memory. When you predict a molecular assembly with AlphaFold-Multimer, the sum of all sequence lengths will have the same limit. So only small assemblies will be possible. To predict longer sequences or larger assemblies you need a GPU with more memory such as an Nvidia A100 with 80 Gbytes, or an A40 with 48 Gbytes. I don't know if Google Colab offers or will offer those in the future. I am currently working on running AlphaFold-Multimer on the UCSF cluster with A40 GPUs to see how well it works and what sequence length it can handle. The memory use is said to increase as the square of the sequence length. We are looking at whether ChimeraX could run AlphaFold predictions on the UCSF Wynton cluster but we probably do not have the budget to pay for that since it would probably cost about $100,000 per year to maintain about 16 high-end GPUs to meet the demand of ChimeraX users. So we are still figuring out what AlphaFold services ChimeraX can offer.
Tom
On Nov 15, 2021, at 9:13 AM, Eric Pettersen via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote:
Not sure why this text does not show up in the text of the message but only in the summary shown by Mail.app, but apparently the text of the message should be:
On Nov 14, 2021, at 9:48 PM, Ichcharisma via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote:
Hi ChimeraX Dev
I wondering if I can run oligomer prediction on ChimeraX AlphaFold features. As we can do it with others colabfold version out there. Thank you _______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chime... https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email.
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
This message from Roden did not make it into the ChimeraX mailing list archive, possibly because our mailing list server was down.
(The original text was with white front color. It shows up for me once the texts are selected... )
Come in to just say that the alphafold2 multimer is still under active debugging. We had faced issue #227 https://github.com/deepmind/alphafold/issues/227 and it is solved days ago. But we are now facing issue #236 https://github.com/deepmind/alphafold/issues/236 frequently, mostly with multimers whose total length exceeds 1000 amino acids. From the discussion, these clashes are also seen in the Colab version. The developer says "We think we may have identified the issue with overlapping domains but need to test a bit more. It will take us a few days at least to be sure." With the current version, it will probably waste a lot of computing time. So one might want to hold the integration to ChimeraX until #236 is solved.
Best, Roden
On Mon, Nov 15, 2021 at 10:32 PM Tom Goddard via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote: ChimeraX does not yet allow running AlphaFold-Multimer the version of AlphaFold released November 2, 2021 that can predict structures for molecular assemblies. We want to offer this, although a ChimeraX version will only be a slight convenience over running an AlphaFold-Multimer Google Colab notebook with a web browser without ChimeraX.
The ChimeraX Google Colab AlphaFold predictions are limited in how long a sequence they can handle to about 800-1000 amino acids. This is because the old Google Colab GPUs (Nvidia K80 and T4) have only 16 Gbytes of memory and AlphaFold runs out of memory. When you predict a molecular assembly with AlphaFold-Multimer, the sum of all sequence lengths will have the same limit. So only small assemblies will be possible. To predict longer sequences or larger assemblies you need a GPU with more memory such as an Nvidia A100 with 80 Gbytes, or an A40 with 48 Gbytes. I don't know if Google Colab offers or will offer those in the future. I am currently working on running AlphaFold-Multimer on the UCSF cluster with A40 GPUs to see how well it works and what sequence length it can handle. The memory use is said to increase as the square of the sequence length. We are looking at whether ChimeraX could run AlphaFold predictions on the UCSF Wynton cluster but we probably do not have the budget to pay for that since it would probably cost about $100,000 per year to maintain about 16 high-end GPUs to meet the demand of ChimeraX users. So we are still figuring out what AlphaFold services ChimeraX can offer.
Tom
On Nov 15, 2021, at 9:13 AM, Eric Pettersen via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote:
Not sure why this text does not show up in the text of the message but only in the summary shown by Mail.app, but apparently the text of the message should be:
On Nov 14, 2021, at 9:48 PM, Ichcharisma via ChimeraX-users <chimerax-users@cgl.ucsf.edu mailto:chimerax-users@cgl.ucsf.edu> wrote:
Hi ChimeraX Dev
I wondering if I can run oligomer prediction on ChimeraX AlphaFold features. As we can do it with others colabfold version out there. Thank you _______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
_______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu mailto:ChimeraX-users@cgl.ucsf.edu Manage subscription: https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chime... https://urldefense.com/v3/__https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users__;!!Nmw4Hv0!m2ihVjg9mTbatGvFzMEJz-7rVpeBmFTWGt_pJIWrnG2pL6EPMIhWk4pR3l515Ymzs-M$
This message and its contents, including attachments are intended solely for the original recipient. If you are not the intended recipient or have received this message in error, please notify me immediately and delete this message from your computer system. Any unauthorized use or distribution is prohibited. Please consider the environment before printing this email._______________________________________________ ChimeraX-users mailing list ChimeraX-users@cgl.ucsf.edu Manage subscription: https://www.rbvi.ucsf.edu/mailman/listinfo/chimerax-users
participants (4)
-
Eric Pettersen -
Ichcharisma -
Roden Deng Luo -
Tom Goddard