2023
DOI: 10.1038/s41592-022-01760-4
|View full text |Cite
|
Sign up to set email alerts
|

Protein structure prediction has reached the single-structure frontier

Abstract: Dramatic advances in protein structure prediction have sparked debate as to whether the problem of predicting structure from sequence is solved or not. Here, I argue that AlphaFold2 and its peers are currently limited by the fact that they predict only a single structure, instead of a structural distribution, and that this realization is crucial for the next generation of structure prediction algorithms.

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

2
60
0

Year Published

2023
2023
2024
2024

Publication Types

Select...
5
3
2

Relationship

0
10

Authors

Journals

citations
Cited by 91 publications
(62 citation statements)
references
References 26 publications
2
60
0
Order By: Relevance
“…As a further indirect validation, a recent experimental work 40 showed that mutation of E 3.39 dramatically enhances the in vitro expression of olfactory receptors, further supporting the structural and functional importance of this residue. As pointed out by a recent commentary, 41 de novo structure determination is dramatically limited by the “single answer problem”: the algorithms return a single structure that is, following the training, the most probable candidate. From a general point of view, this can be correct only in the realm of single-state proteins, while here (and in the majority of the biologically-relevant cases) our target GPCR has a set of different conformational states.…”
Section: Models With Sodium In Its Binding Sitementioning
confidence: 99%
“…As a further indirect validation, a recent experimental work 40 showed that mutation of E 3.39 dramatically enhances the in vitro expression of olfactory receptors, further supporting the structural and functional importance of this residue. As pointed out by a recent commentary, 41 de novo structure determination is dramatically limited by the “single answer problem”: the algorithms return a single structure that is, following the training, the most probable candidate. From a general point of view, this can be correct only in the realm of single-state proteins, while here (and in the majority of the biologically-relevant cases) our target GPCR has a set of different conformational states.…”
Section: Models With Sodium In Its Binding Sitementioning
confidence: 99%
“…Furthermore, the initial predictions of the native structures of proteins have been recently extended to protein complexes 4,5 . These advances have prompted the question of whether it is possible to use this type of approach for the prediction of the conformational fluctuations of the native states of folded proteins [6][7][8][9][10][11][12] , and more generally for the characterisation of the structural properties of the native states of disordered proteins [13][14][15] . Support for this idea comes from the observation that AlphaFold performs as well as current state-of-the-art predictors of protein disorder 16,17 .…”
Section: Introductionmentioning
confidence: 99%
“…One important research direction is to extend FrameDiff to conditional generative modeling tasks, such as probabilistic sequence-to-structure prediction which is key to capture functional motion (Lane, 2023), but also probabilistic scaffolding design given a functional motif (Trippe et al, 2022) In terms of methodology, we stress the importance of understanding and questioning the benefits and necessity of pretraining with protein structure prediction which heavily relies on evolutionary couplings that are unused in de novo protein design. Given our preliminary results, we hypothesize scaling up FrameDiff to train with more data and improvements in sampling designable backbones could overcome the need for pretraining with protein structure prediction.…”
Section: Discussionmentioning
confidence: 99%