A structural biology community assessment of AlphaFold2 applications

By Mehmet Akdel, Douglas E. V. Pires, Eduard Porta Pardo, Jürgen Jänes, Arthur O. Zalevsky, Bálint Mészáros, Patrick Bryant, Lydia L. Good, Roman A. Laskowski, Gabriele Pozzati and 21 more. 2021

578 citations recorded by the source corpus.

Most proteins fold into 3D structures that determine how they function and orchestrate the biological processes of the cell. Recent developments in computational methods have led to protein structure predictions that have reached the accuracy of experimentally determined models. While this has been independently verified, the implementation of these methods across structural biology applications remains to be tested. Here, we evaluate the use of AlphaFold 2 (AF2) predictions in the study of characteristic structural elements; the impact of missense variants; function and ligand binding site predictions; modelling of interactions; and modelling of experimental structural data. For 11 proteomes, an average of 25% additional residues can be confidently modelled when compared to homology modelling, identifying structural features rarely seen in the PDB. AF2-based predictions of protein disorder and protein complexes surpass state-of-the-art tools and AF2 models can be used across diverse applications equally well compared to experimentally determined structures, when the confidence metrics are critically considered. In summary, we find that these advances are likely to have a transformative impact in structural biology and broader life science research.

Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Theory and Mathematics

In Materials Chemistry

In Molecular Biology

Citations in the corpus

AlphaFold2 and its applications in the fields of biology and medicine (2023, background): “65 Rather than saying that AF2 may make structural biologists unemployed, we prefer to the viewpoint that the AF2 and its predicted structures will change the way we do structural biology, including X-ray crystallography, cryo-EM, and NMR spectroscopy.”
Protein Representation Learning by Geometric Structure Pretraining (2022, methodology): “Following Akdel et al. (2021), we highlight the 20 most common super-26”
Systematic evaluation of computational tools to predict the effects of mutations on protein stability in the absence of experimental structures (2022, background): “In addition to template-based approaches for protein structure prediction, more recently, the development of AlphaFold2 [62] has revolutionized the field with a significant increase in performance, promising to bridge the gap in protein 3D information with high-quality models [63].”
Computational approaches streamlining drug discovery (2023, background): “99,100) or RosettaFold 101 3D models, which already show utility in many applications, including protein–protein and protein–peptide docking 102 .”
AlphaFold2 models indicate that protein sequence determines both structure and dynamics (2022, background): “High pLDDT scores (e.g., > 80) indicate high confidence of the residue structure, and low pLDDT scores (e.g., 50) may indicate that the residues are in intrinsically disordered protein regions (IDPRs) 19,31,32 .”
AlphaFold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination (2023, unknown): “…predictions vary substantially in their global and local agreement with deposited models and also in their coverage at the highest levels of confidence 1,11,21 , with only 36% of residues in the human proteome 22 and 73% of residues in Escherichia coli modeled with very high confidence 23 .”
AlphaFold2 can predict structural and phenotypic effects of single mutations (2022, background): “14,15 In summary, we showed here that AF2 predictions of local structural change (without extra machine-learning algorithms), can be used to study missense mutations in proteins.”
The impact of AlphaFold2 one year on (2022, unknown): “Before long there may be a whole family of methods, some of which may be better (on average) than the current version of AlphaFold, but there will then be no obvious way to pick just one ‘best’ model from the alternatives available.”
The prospects and opportunities of protein structure prediction with AI (2022, methodology): “For X-ray crystallographers, the routine availability of good models has enabled more structures to be solved by molecular replacement, reducing the time spent on experimental phasing 2 .”
AlphaDesign: A de novo protein design framework based on AlphaFold (2021, methodology): “Whilst intended for single-chain structure prediction, an unexpected consequence of AF’s input protocol allows prediction of non-contiguous chains, thus enabling prediction of protein complexes using the existing trained network [18, 19].”
Harnessing protein folding neural networks for peptide–protein docking (2021, background): “The latter has been implemented for the modeling of homo-and hetero-multimers in several recent studies on AF2 36,37 .”
Complementing machine learning‐based structure predictions with native mass spectrometry (2022, background): “Furthermore, using AF2 to predict interactions involving dynamic regions,(6) ligand binding sites, or point mutations,(7) all of which are major focal points of structural biology, remains challenging.(8) In these cases, additional structural data may be required to assess the validity of the comput”
Clustering predicted structures at the scale of the known protein universe (2023, methodology): “We predicted small-molecule-binding sites for representative dark cluster members by adapting a previously described approach 9 .”
AI-based structure prediction empowers integrative structural analysis of human nuclear pores (2022, background): “However, AlphaFold and RoseTTAfold have recently demonstrated unprecedented accuracy in predicting structures ofmonomeric proteins (26, 27, 60–65) and complexes (34, 36, 61, 66).”
Modeling Alternate Conformations with Alphafold2 via Modification of the Multiple Sequence Alignment (2021, methodology): “Recent reports have utilized AF2 models 291 and existing paradigms to ascertain the potential effect of a mutation without a discernible 292 correlation.(24,25) Here the structures of the mutant protein lead to an increase in disorder or altered 293 conformations.”