AlphaFold 3 Accuracy on Antibody Binding and Protein Interactions

When is AlphaFold 3 wrong? How can we tell? What should we do? This
video discusses cases where AlphaFold 3 was incorrect on our proteins.
Similar to recent posts on generative AI from many colleagues, an analysis of protein structure predictions that AlphaFold3 got wrong for our proteins, with speculations on why.
With the remarkable success of the AlphaFold prediction yielding
accurate structural models (Abramson et al., 2024), studies based
entirely on AlphaFold analysis have gained popularity. It is however
important to note that AlphaFold is fundamentally a statistical method
relying on existing experimental structures in the PDB database, and
it is not based on biophysical principles. Hence, its prediction can
be biased by existing experimental structures. The cases presented
here are a sober reminder that despite the inspiring "structures"
generated by AlphaFold, they remain as computer models and require
rigorous experimental investigation to unearth the structural basis
underlying important biological processes. (Paraphrased from Forker et
al., 2024).
Here, our structural elucidation (Holt et al., 2023) of how designed
antibodies bind to HIV Env (envelope) protein provides molecular
evidence differing drastically from the AlphaFold modeling. The
correct structures will accelerate the development of anti-HIV antibodies
and vaccines.
References:
Holt, G. et al. Cell Reports (2023)
Improved HIV-1 neutralization breadth and potency of V2-apex antibodies by in silico design
42(7):112711.
doi: 10.1016/j.celrep.2023.112711.
www.sciencedir...
(See SR 6: ars.els-cdn.co... )
Abramson, J. et al. Nature (2024)
www.nature.com...
Forker, K. et al. EMBO J. (2024). Crystal structure of MAGEA4 MHD-RAD18 R6BD reveals a flipped binding mode compared to AlphaFold prediction. doi: 10.1038/s44318-024-00140-2. www.embopress....
#ai #proteinstructure #machinelearning

[@Scientific_Hustler]

Professor, your presentation was insightful. It reminded me of a quote by Professor Henry that says *"A model must be wrong, in some respects, else it would be the thing itself. The trick is to see where it is right"*

[@soremekunoladimeji5289]

Thank you so much Professor. I learnt understanding the subtle yet significant errors in computational protein modeling is essential for researchers pursuing drug discovery, as it can greatly inform and improve their research endeavors.

[@lisizecha9759]

Thanks for sharing this video

[@DonaldLab]

My pleasure