Hi, This question isn’t specifically about chimerax, but I hope it’s in the ball park. For most of the protein pairs I’ve tested using multimer, the ipTM scores for the 5-10 models are relatively consistent. For example all 0.2–0.3, or 0.75–0.85, or 0.4–0.6. But occasionally I’ll get one or two very high scores (>0.7) with the rest being low or somewhat randomly scattered. My instinct, of course, is to put more faith in scores when the models converge. At the same time, I don’t want ignore a potentially real interaction. So, I’m wondering how people interpret ipTMs when the numbers are variable between models. What is the reason that some protein pairs show this pattern? Is the good score(s) indicative or something real or a false positive. Are the bad scores indicative of true or false negative? Any thoughts, insights, and possible follow up approaches would be greatly appreciated! David
Hi David, Other than feedback from other ChimeraX users (anybody?)... Maybe you can ask the AlphaFold-Multimer developers, or maybe their publication has more statistics on this kind of thing. Protein complex prediction with AlphaFold-Multimer. Evans R, O'Neill M, Pritzel A, et al. bioRxiv 2021. <https://www.biorxiv.org/content/10.1101/2021.10.04.463034v2> Other than that, all I can suggest is to take a look at the following links for some discussion and ChimeraX tools for AlphaFold dimer prediction. However, we are not the authorities on this topic. <https://www.rbvi.ucsf.edu/chimerax/data/alphapairs-oct-2023/alphapairs.html> <https://www.rbvi.ucsf.edu/chimerax/data/afbatch-jan2024/rim_dimers.html> I hope this helps, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Resource for Biocomputing, Visualization, and Informatics Department of Pharmaceutical Chemistry University of California, San Francisco
On Mar 14, 2024, at 5:55 AM, David S. Fay via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Hi,
This question isn’t specifically about chimerax, but I hope it’s in the ball park. For most of the protein pairs I’ve tested using multimer, the ipTM scores for the 5-10 models are relatively consistent. For example all 0.2–0.3, or 0.75–0.85, or 0.4–0.6. But occasionally I’ll get one or two very high scores (>0.7) with the rest being low or somewhat randomly scattered. My instinct, of course, is to put more faith in scores when the models converge. At the same time, I don’t want ignore a potentially real interaction. So, I’m wondering how people interpret ipTMs when the numbers are variable between models. What is the reason that some protein pairs show this pattern? Is the good score(s) indicative or something real or a false positive. Are the bad scores indicative of true or false negative?
Any thoughts, insights, and possible follow up approaches would be greatly appreciated!
David
Hi David, I also sometimes see Alphafold produce 5 predictions of a multimer with a few having high confidence and a few having low confidence. It is not too common, maybe 5% of multimers. I think the high confidence predictions are very likely correct in these cases. I have done predictions of hundreds of different multimers and have never seen Alphafold give a high confidence prediction that definitely has wrong binding interfaces. I would love to see examples of this false positive binding scenario. On the other hand Alphafold often fails to find real multimer interfaces. I recall the Alphafold multimer paper benchmark on thousands of PDB experimental models failed to find 1/3 of the binding interfaces. That agrees with my experience. False negatives are very common. Tom
On Mar 15, 2024, at 7:57 AM, Elaine Meng via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Hi David, Other than feedback from other ChimeraX users (anybody?)...
Maybe you can ask the AlphaFold-Multimer developers, or maybe their publication has more statistics on this kind of thing.
Protein complex prediction with AlphaFold-Multimer. Evans R, O'Neill M, Pritzel A, et al. bioRxiv 2021. <https://www.biorxiv.org/content/10.1101/2021.10.04.463034v2>
Other than that, all I can suggest is to take a look at the following links for some discussion and ChimeraX tools for AlphaFold dimer prediction. However, we are not the authorities on this topic.
<https://www.rbvi.ucsf.edu/chimerax/data/alphapairs-oct-2023/alphapairs.html> <https://www.rbvi.ucsf.edu/chimerax/data/afbatch-jan2024/rim_dimers.html>
I hope this helps, Elaine ----- Elaine C. Meng, Ph.D. UCSF Chimera(X) team Resource for Biocomputing, Visualization, and Informatics Department of Pharmaceutical Chemistry University of California, San Francisco
On Mar 14, 2024, at 5:55 AM, David S. Fay via ChimeraX-users <chimerax-users@cgl.ucsf.edu> wrote:
Hi,
This question isn’t specifically about chimerax, but I hope it’s in the ball park. For most of the protein pairs I’ve tested using multimer, the ipTM scores for the 5-10 models are relatively consistent. For example all 0.2–0.3, or 0.75–0.85, or 0.4–0.6. But occasionally I’ll get one or two very high scores (>0.7) with the rest being low or somewhat randomly scattered. My instinct, of course, is to put more faith in scores when the models converge. At the same time, I don’t want ignore a potentially real interaction. So, I’m wondering how people interpret ipTMs when the numbers are variable between models. What is the reason that some protein pairs show this pattern? Is the good score(s) indicative or something real or a false positive. Are the bad scores indicative of true or false negative?
Any thoughts, insights, and possible follow up approaches would be greatly appreciated!
David
_______________________________________________ ChimeraX-users mailing list -- chimerax-users@cgl.ucsf.edu To unsubscribe send an email to chimerax-users-leave@cgl.ucsf.edu Archives: https://mail.cgl.ucsf.edu/mailman/archives/list/chimerax-users@cgl.ucsf.edu/
participants (3)
-
David S. Fay -
Elaine Meng -
Tom Goddard