Super cool work! I’m a little confused though. The article starts with saying the challenge with designing enzymes is that the crystal structures don’t represent the required dynamics needed for catalysis. It feels like this new approach was presented as a way to overcome that, but doesn’t the fact that they can get crystal structures for all the intermediate states which match the diffusion output show this wasn’t an issue? Maybe I’m missing something but if the dynamics can be represented by static crystal structures then the initial challenge mentioned doesn’t seem to apply here. Or was one of the goals to design enzymes with more resolvable dynamics?
So, the paper authors don't actually grab crystal structures of all intermediate states. From the paper:
“””
We pursued x-ray crystallography to determine the accuracy with which super and win were designed. We were able to solve crystal structures of both super and win, and found that they had very low Cα RMSDs of 0.8 Å οver 165 residues and 0.83 Å over 160 residues (Fig. 3A,D), respectively, to the design models....Although the structures were solved in the absence of bound small molecule substrate or transition state
analogue, overlay of the design model and crystal structure of super reveals high shape complementarity to the butyrate acyl group of its preferred substrate.
“””
The structures mainly show that the designed scaffolds match up with how the scaffolds fold in real life! It cannot necessarily check that the steps of the catalysis process of the designed enzyme match up with how real, natural serine hydrolyses work. That comes from functional assays!
Hope this first step helps lead to not just new enzymes, but also new catalytic activities in the future! The big missing link appears to be the inability to predict actual chemical/molecular dynamics. How computationally intensive would be to simulate that multi-step reaction for the top designs that pass the earlier filters? It's incredible that their crystal structures matched predictions so well for the successful designs!
(Thanks for the great summary - it'll be very interesting to learn from those failure mode analyses and maybe use that information to train future ML models)
This represents a shift in medical research and innovation away from discovery and towards engineering. Such a shift has huge implications both for the speed of innovation and the need to adapt and accept new approaches to regulation.
thanks for the great summary, just letting you know I think there’s a typo where you say Cas9 is commonly used in the dairy industry? it would please me greatly if we were CRISPRing all the cows but i think you meant chymosin.
Super cool work! I’m a little confused though. The article starts with saying the challenge with designing enzymes is that the crystal structures don’t represent the required dynamics needed for catalysis. It feels like this new approach was presented as a way to overcome that, but doesn’t the fact that they can get crystal structures for all the intermediate states which match the diffusion output show this wasn’t an issue? Maybe I’m missing something but if the dynamics can be represented by static crystal structures then the initial challenge mentioned doesn’t seem to apply here. Or was one of the goals to design enzymes with more resolvable dynamics?
Hi, one of the authors of the essay here!
So, the paper authors don't actually grab crystal structures of all intermediate states. From the paper:
“””
We pursued x-ray crystallography to determine the accuracy with which super and win were designed. We were able to solve crystal structures of both super and win, and found that they had very low Cα RMSDs of 0.8 Å οver 165 residues and 0.83 Å over 160 residues (Fig. 3A,D), respectively, to the design models....Although the structures were solved in the absence of bound small molecule substrate or transition state
analogue, overlay of the design model and crystal structure of super reveals high shape complementarity to the butyrate acyl group of its preferred substrate.
“””
The structures mainly show that the designed scaffolds match up with how the scaffolds fold in real life! It cannot necessarily check that the steps of the catalysis process of the designed enzyme match up with how real, natural serine hydrolyses work. That comes from functional assays!
Hope this first step helps lead to not just new enzymes, but also new catalytic activities in the future! The big missing link appears to be the inability to predict actual chemical/molecular dynamics. How computationally intensive would be to simulate that multi-step reaction for the top designs that pass the earlier filters? It's incredible that their crystal structures matched predictions so well for the successful designs!
(Thanks for the great summary - it'll be very interesting to learn from those failure mode analyses and maybe use that information to train future ML models)
Catalytic triad is one of the (few) cool monikers in biochemistry, how did you not use it?!? 😭
Great read
This represents a shift in medical research and innovation away from discovery and towards engineering. Such a shift has huge implications both for the speed of innovation and the need to adapt and accept new approaches to regulation.
thanks for the great summary, just letting you know I think there’s a typo where you say Cas9 is commonly used in the dairy industry? it would please me greatly if we were CRISPRing all the cows but i think you meant chymosin.