Example table: protein structure and design

Créer un compte gratuit pour copier ces articles Se connecter

Articles

Highly accurate protein structure prediction with AlphaFold.
Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, Tunyasuvunakool K, Bates R · DOI: 10.1038/s41586-021-03819-2 · 2021 · Nature

Publication retrouvée
Accurate structure prediction of biomolecular interactions with AlphaFold 3
Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore · DOI: 10.1038/s41586-024-07487-w · 2024 · Nature
De novo design of protein structure and function with RFdiffusion
Joseph L. Watson, David Juergens, Nathaniel R. Bennett, Brian L. Trippe, Jason Yim, Helen E. Eisenach, Woody Ahern, Andrew J. Borst · DOI: 10.1038/s41586-023-06415-8 · 2023 · Nature
Illuminating protein space with a programmable generative model
John Ingraham, Max Baranov, Zak Costello, Karl W. Barber, Wujie Wang, Ahmed Ismail, Vincent Frappier, Dana M. Lord · DOI: 10.1038/s41586-023-06728-8 · 2023 · Nature
ColabFold: making protein folding accessible to all
Milot Mirdita, Konstantin Schütze, Yoshitaka Moriwaki, Lim Heo, Sergey Ovchinnikov, Martin Steinegger · DOI: 10.1038/s41592-022-01488-1 · 2022 · Nature Methods

Tableau comparatif

18 cases remplies sur 18 s’appuient sur un passage de l’article.

Article	Question	Méthode	Données	Résultats principaux	Limites
1 Highly accurate protein structure prediction with AlphaFold.	Développer une méthode capable de prédire régulièrement des structures protéiques avec une précision atomique, même sans structure similaire connue. Source · Abstract Here we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known.	AlphaFold combine apprentissage profond, connaissances physiques et biologiques et alignements multiples de séquences. Une auto-distillation améliore l’apprentissage supervisé sur les structures de la PDB. Source · Training with labelled and unlabelled d… The AlphaFold architecture is able to train to high accuracy using only supervised learning on PDB data, but we are able to enhance accuracy (Fig. 4a) using an approach similar to noisy student self-distillation35.	L’apprentissage associe les données de la PDB à un sous-ensemble de structures prédites à haute confiance, issu d’environ 350 000 séquences diverses d’Uniclust30. Source · Training with labelled and unlabelled d… In this procedure, we use a trained network to predict the structure of around 350,000 diverse sequences from Uniclust3036 and make a new dataset of predicted structures filtered to a high-confidence subset. We then train the same architecture again from scratch using a mixture of PDB data and this new dataset of predicted structures as the training data, in which the various training data augmentations such as cropping and MSA subsampling make it challenging for the network to recapitulate the previously predicted structures.	Lors de CASP14, AlphaFold atteint une précision comparable aux structures expérimentales dans la majorité des cas et surpasse nettement les autres méthodes. L’extrait ne fournit pas de score numérique de précision. Source · Abstract We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14) 15 , demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods.	Non précisé
2 Accurate structure prediction of biomolecular interactions with AlphaFold 3 Résumé d’auteur et notre synthèse	Prédire dans un cadre unifié la structure de complexes associant protéines, acides nucléiques, petites molécules, ions et résidus modifiés. Source · Abstract Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.	AlphaFold 3 utilise une architecture profondément remaniée, fondée sur la diffusion, pour prédire conjointement les structures des composants d’un complexe. Source · Abstract Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.	Non précisé	La précision dépasse celle des outils spécialisés comparés pour les interactions protéine–ligand et protéine–acide nucléique, ainsi que celle d’AlphaFold-Multimer v.2.3 pour les anticorps–antigènes. Aucun score chiffré n’est fourni dans l’extrait. Source · Abstract The new AlphaFold model demonstrates substantially improved accuracy over many previous specialized tools: far greater accuracy for protein–ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein–nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody–antigen prediction accuracy compared with AlphaFold-Multimer v.2.3 7,8 .	Non précisé
3 De novo design of protein structure and function with RFdiffusion Résumé d’auteur et notre synthèse	Créer un cadre génératif polyvalent pour concevoir des protéines fonctionnelles à partir de spécifications moléculaires simples. Source · Abstract In a manner analogous to networks that produce images from user-specified inputs, RFdiffusion enables the design of diverse functional proteins from simple molecular specifications.	RFdiffusion est obtenu en affinant le réseau RoseTTAFold sur des tâches de débruitage structural. Il génère des squelettes protéiques pour la conception de monomères, de partenaires de liaison, d’oligomères symétriques et de structures portant des motifs fonctionnels. Source · Abstract Here we show that by fine-tuning the RoseTTAFold structure prediction network on protein structure denoising tasks, we obtain a generative model of protein backbones that achieves outstanding performance on unconditional and topology-constrained protein monomer design, protein binder design, symmetric oligomer design, enzyme active site scaffolding and symmetric motif scaffolding for therapeutic and metal-binding protein design.	La validation expérimentale porte sur des centaines d’assemblages symétriques, de protéines liant des métaux et de partenaires de liaison conçus. Le nombre exact n’est pas indiqué. Source · Abstract We demonstrate the power and generality of the method, called RoseTTAFold diffusion (RFdiffusion), by experimentally characterizing the structures and functions of hundreds of designed symmetric assemblies, metal-binding proteins and protein binders.	Une structure obtenue par cryomicroscopie électronique d’un partenaire conçu en complexe avec l’hémagglutinine grippale est presque identique au modèle de conception. Aucun écart structural chiffré n’est fourni. Source · Abstract The accuracy of RFdiffusion is confirmed by the cryogenic electron microscopy structure of a designed binder in complex with influenza haemagglutinin that is nearly identical to the design model.	Non précisé
4 Illuminating protein space with a programmable generative model Résumé d’auteur et notre synthèse	Développer un modèle génératif programmable produisant de nouvelles structures et séquences de protéines, orientées vers des propriétés ou fonctions souhaitées. Source · Abstract Here we introduce Chroma, a generative model for proteins and protein complexes that can directly sample novel protein structures and sequences, and that can be conditioned to steer the generative process towards desired properties and functions.	Chroma combine une diffusion respectant les statistiques conformationnelles des polymères et une architecture à complexité sous-quadratique. La conception est formulée comme une inférence bayésienne sous contraintes, notamment de symétrie, de forme ou de langage naturel. Source · Abstract To enable this, we introduce a diffusion process that respects the conformational statistics of polymer ensembles, an efficient neural architecture for molecular systems that enables long-range reasoning with sub-quadratic scaling, layers for efficiently synthesizing three-dimensional structures of proteins from predicted inter-residue geometries and a general low-temperature sampling algorithm for diffusion models. Chroma achieves protein design as Bayesian inference under external constraints, which can involve symmetries, substructure, shape, semantics and even natural-language prompts.	Les auteurs caractérisent expérimentalement 310 protéines générées et déterminent les structures cristallines de deux protéines conçues. Source · Abstract The experimental characterization of 310 proteins shows that sampling from Chroma results in proteins that are highly expressed, fold and have favourable biophysical properties. The crystal structures of two designed proteins exhibit atomistic agreement with Chroma samples (a backbone root-mean-square deviation of around 1.0 Å).	Les protéines caractérisées présentent une forte expression, se replient et possèdent des propriétés biophysiques favorables. Pour deux structures cristallines, l’écart quadratique moyen du squelette avec les modèles générés est d’environ 1,0 Å. Source · Abstract The experimental characterization of 310 proteins shows that sampling from Chroma results in proteins that are highly expressed, fold and have favourable biophysical properties. The crystal structures of two designed proteins exhibit atomistic agreement with Chroma samples (a backbone root-mean-square deviation of around 1.0 Å).	Non précisé
5 ColabFold: making protein folding accessible to all Résumé d’auteur et notre synthèse	Accélérer la prédiction des structures de protéines et de complexes et la rendre accessible gratuitement via Google Colaboratory. Source · Abstract ColabFold offers accelerated prediction of protein structures and complexes by combining the fast homology search of MMseqs2 with AlphaFold2 or RoseTTAFold. ColabFold's 40-60-fold faster search and optimized model utilization enables prediction of close to 1,000 structures per day on a server with one graphics processing unit. Coupled with Google Colaboratory, ColabFold becomes a free and accessible platform for protein folding.	ColabFold associe la recherche rapide d’homologues par MMseqs2 à AlphaFold2 ou RoseTTAFold. Source · Abstract ColabFold offers accelerated prediction of protein structures and complexes by combining the fast homology search of MMseqs2 with AlphaFold2 or RoseTTAFold.	Non précisé	La recherche est accélérée d’un facteur 40 à 60. L’optimisation permet de prédire près de 1 000 structures par jour sur un serveur équipé d’un seul processeur graphique. Source · Abstract ColabFold's 40-60-fold faster search and optimized model utilization enables prediction of close to 1,000 structures per day on a server with one graphics processing unit.	Non précisé

Où ils s'accordent

Les deux générations d’AlphaFold visent une prédiction structurale précise par apprentissage profond. Leurs résultats indiquent des progrès importants par rapport aux méthodes auxquelles elles sont comparées, sans score numérique de précision dans le tableau. [1, 2]
6 passages derrière ce constat

1 · Question (Abstract): Here we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known.
1 · Méthode (Training with labelled and unlabelled data): The AlphaFold architecture is able to train to high accuracy using only supervised learning on PDB data, but we are able to enhance accuracy (Fig. 4a) using an approach similar to noisy student self-distillation35.
1 · Résultats principaux (Abstract): We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14) 15 , demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods.
2 · Question (Abstract): Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.
2 · Méthode (Abstract): Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.
2 · Résultats principaux (Abstract): The new AlphaFold model demonstrates substantially improved accuracy over many previous specialized tools: far greater accuracy for protein–ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein–nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody–antigen prediction accuracy compared with AlphaFold-Multimer v.2.3 7,8 .
AlphaFold 3, RFdiffusion et Chroma emploient la diffusion pour modéliser des structures moléculaires, mais avec des finalités différentes : prédire des complexes ou concevoir de nouvelles protéines. [2, 3, 4]
6 passages derrière ce constat

2 · Question (Abstract): Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.
2 · Méthode (Abstract): Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.
3 · Question (Abstract): In a manner analogous to networks that produce images from user-specified inputs, RFdiffusion enables the design of diverse functional proteins from simple molecular specifications.
3 · Méthode (Abstract): Here we show that by fine-tuning the RoseTTAFold structure prediction network on protein structure denoising tasks, we obtain a generative model of protein backbones that achieves outstanding performance on unconditional and topology-constrained protein monomer design, protein binder design, symmetric oligomer design, enzyme active site scaffolding and symmetric motif scaffolding for therapeutic and metal-binding protein design.
4 · Question (Abstract): Here we introduce Chroma, a generative model for proteins and protein complexes that can directly sample novel protein structures and sequences, and that can be conditioned to steer the generative process towards desired properties and functions.
4 · Méthode (Abstract): To enable this, we introduce a diffusion process that respects the conformational statistics of polymer ensembles, an efficient neural architecture for molecular systems that enables long-range reasoning with sub-quadratic scaling, layers for efficiently synthesizing three-dimensional structures of proteins from predicted inter-residue geometries and a general low-temperature sampling algorithm for diffusion models. Chroma achieves protein design as Bayesian inference under external constraints, which can involve symmetries, substructure, shape, semantics and even natural-language prompts.
RFdiffusion et Chroma permettent une conception générative guidée par des spécifications, notamment de symétrie, et rapportent une validation expérimentale de protéines conçues. [3, 4]
8 passages derrière ce constat

3 · Question (Abstract): In a manner analogous to networks that produce images from user-specified inputs, RFdiffusion enables the design of diverse functional proteins from simple molecular specifications.
3 · Méthode (Abstract): Here we show that by fine-tuning the RoseTTAFold structure prediction network on protein structure denoising tasks, we obtain a generative model of protein backbones that achieves outstanding performance on unconditional and topology-constrained protein monomer design, protein binder design, symmetric oligomer design, enzyme active site scaffolding and symmetric motif scaffolding for therapeutic and metal-binding protein design.
3 · Données (Abstract): We demonstrate the power and generality of the method, called RoseTTAFold diffusion (RFdiffusion), by experimentally characterizing the structures and functions of hundreds of designed symmetric assemblies, metal-binding proteins and protein binders.
3 · Résultats principaux (Abstract): The accuracy of RFdiffusion is confirmed by the cryogenic electron microscopy structure of a designed binder in complex with influenza haemagglutinin that is nearly identical to the design model.
4 · Question (Abstract): Here we introduce Chroma, a generative model for proteins and protein complexes that can directly sample novel protein structures and sequences, and that can be conditioned to steer the generative process towards desired properties and functions.
4 · Méthode (Abstract): To enable this, we introduce a diffusion process that respects the conformational statistics of polymer ensembles, an efficient neural architecture for molecular systems that enables long-range reasoning with sub-quadratic scaling, layers for efficiently synthesizing three-dimensional structures of proteins from predicted inter-residue geometries and a general low-temperature sampling algorithm for diffusion models. Chroma achieves protein design as Bayesian inference under external constraints, which can involve symmetries, substructure, shape, semantics and even natural-language prompts.
4 · Données (Abstract): The experimental characterization of 310 proteins shows that sampling from Chroma results in proteins that are highly expressed, fold and have favourable biophysical properties. The crystal structures of two designed proteins exhibit atomistic agreement with Chroma samples (a backbone root-mean-square deviation of around 1.0 Å).
4 · Résultats principaux (Abstract): The experimental characterization of 310 proteins shows that sampling from Chroma results in proteins that are highly expressed, fold and have favourable biophysical properties. The crystal structures of two designed proteins exhibit atomistic agreement with Chroma samples (a backbone root-mean-square deviation of around 1.0 Å).

Où ils divergent

Le tableau ne permet pas d’identifier de contradiction directe entre les résultats : les objectifs, les évaluations et les indicateurs rapportés diffèrent. Aucune comparaison quantitative commune aux cinq travaux n’est fournie. [1, 2, 3, 4, 5]
10 passages derrière ce constat

1 · Question (Abstract): Here we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known.
1 · Résultats principaux (Abstract): We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14) 15 , demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods.
2 · Question (Abstract): Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.
2 · Résultats principaux (Abstract): The new AlphaFold model demonstrates substantially improved accuracy over many previous specialized tools: far greater accuracy for protein–ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein–nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody–antigen prediction accuracy compared with AlphaFold-Multimer v.2.3 7,8 .
3 · Question (Abstract): In a manner analogous to networks that produce images from user-specified inputs, RFdiffusion enables the design of diverse functional proteins from simple molecular specifications.
3 · Résultats principaux (Abstract): The accuracy of RFdiffusion is confirmed by the cryogenic electron microscopy structure of a designed binder in complex with influenza haemagglutinin that is nearly identical to the design model.
4 · Question (Abstract): Here we introduce Chroma, a generative model for proteins and protein complexes that can directly sample novel protein structures and sequences, and that can be conditioned to steer the generative process towards desired properties and functions.
4 · Résultats principaux (Abstract): The experimental characterization of 310 proteins shows that sampling from Chroma results in proteins that are highly expressed, fold and have favourable biophysical properties. The crystal structures of two designed proteins exhibit atomistic agreement with Chroma samples (a backbone root-mean-square deviation of around 1.0 Å).
5 · Question (Abstract): ColabFold offers accelerated prediction of protein structures and complexes by combining the fast homology search of MMseqs2 with AlphaFold2 or RoseTTAFold. ColabFold's 40-60-fold faster search and optimized model utilization enables prediction of close to 1,000 structures per day on a server with one graphics processing unit. Coupled with Google Colaboratory, ColabFold becomes a free and accessible platform for protein folding.
5 · Résultats principaux (Abstract): ColabFold's 40-60-fold faster search and optimized model utilization enables prediction of close to 1,000 structures per day on a server with one graphics processing unit.
Le périmètre de prédiction s’élargit entre les deux générations d’AlphaFold : la première vise les structures protéiques, y compris sans structure similaire connue, tandis que la troisième traite conjointement protéines, acides nucléiques, petites molécules, ions et résidus modifiés. Leurs supériorités respectives ne portent donc pas sur les mêmes comparaisons. [1, 2]
4 passages derrière ce constat

1 · Question (Abstract): Here we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known.
1 · Résultats principaux (Abstract): We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14) 15 , demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods.
2 · Question (Abstract): Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.
2 · Résultats principaux (Abstract): The new AlphaFold model demonstrates substantially improved accuracy over many previous specialized tools: far greater accuracy for protein–ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein–nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody–antigen prediction accuracy compared with AlphaFold-Multimer v.2.3 7,8 .
Les validations des deux modèles de conception ne sont pas rapportées de la même façon. RFdiffusion est évalué sur des centaines de réalisations expérimentales, sans effectif exact, et une structure en complexe est décrite comme presque identique au modèle sans écart chiffré. Chroma caractérise 310 protéines et rapporte, pour deux structures cristallines, un écart quadratique moyen du squelette d’environ 1,0 Å. Ces données ne suffisent pas à classer leur précision. [3, 4]
4 passages derrière ce constat

3 · Données (Abstract): We demonstrate the power and generality of the method, called RoseTTAFold diffusion (RFdiffusion), by experimentally characterizing the structures and functions of hundreds of designed symmetric assemblies, metal-binding proteins and protein binders.
3 · Résultats principaux (Abstract): The accuracy of RFdiffusion is confirmed by the cryogenic electron microscopy structure of a designed binder in complex with influenza haemagglutinin that is nearly identical to the design model.
4 · Données (Abstract): The experimental characterization of 310 proteins shows that sampling from Chroma results in proteins that are highly expressed, fold and have favourable biophysical properties. The crystal structures of two designed proteins exhibit atomistic agreement with Chroma samples (a backbone root-mean-square deviation of around 1.0 Å).
4 · Résultats principaux (Abstract): The experimental characterization of 310 proteins shows that sampling from Chroma results in proteins that are highly expressed, fold and have favourable biophysical properties. The crystal structures of two designed proteins exhibit atomistic agreement with Chroma samples (a backbone root-mean-square deviation of around 1.0 Å).
ColabFold rapporte surtout un gain d’efficacité : recherche d’homologues accélérée d’un facteur 40 à 60 et près de 1 000 structures prédites par jour sur un serveur à un seul processeur graphique. À l’inverse, les deux articles AlphaFold mettent en avant la précision, sans score numérique dans le tableau. Il s’agit d’indicateurs différents, et non de résultats contradictoires. [1, 2, 5]
6 passages derrière ce constat

5 · Méthode (Abstract): ColabFold offers accelerated prediction of protein structures and complexes by combining the fast homology search of MMseqs2 with AlphaFold2 or RoseTTAFold.
5 · Résultats principaux (Abstract): ColabFold's 40-60-fold faster search and optimized model utilization enables prediction of close to 1,000 structures per day on a server with one graphics processing unit.
1 · Méthode (Training with labelled and unlabelled data): The AlphaFold architecture is able to train to high accuracy using only supervised learning on PDB data, but we are able to enhance accuracy (Fig. 4a) using an approach similar to noisy student self-distillation35.
1 · Résultats principaux (Abstract): We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14) 15 , demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods.
2 · Méthode (Abstract): Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues.
2 · Résultats principaux (Abstract): The new AlphaFold model demonstrates substantially improved accuracy over many previous specialized tools: far greater accuracy for protein–ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein–nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody–antigen prediction accuracy compared with AlphaFold-Multimer v.2.3 7,8 .

Ce qu'aucun n'aborde

Quels sont les modes d’échec, les limites de validité et les conditions où chaque méthode devient peu fiable ? Aucune limitation n’est précisée dans le tableau.
Quels compromis entre précision, temps de calcul, mémoire et consommation énergétique apparaîtraient dans une évaluation sur des tâches et un matériel communs ? Le tableau ne fournit pas une telle comparaison.
Dans quelle mesure les modèles décrivent-ils la dynamique moléculaire, les états conformationnels alternatifs et leurs populations, plutôt qu’une structure cible ? Cette question n’est pas abordée dans les informations fournies.
Comment les performances varient-elles selon la distance aux données d’apprentissage, et comment les biais ou les recouvrements entre entraînement et évaluation sont-ils contrôlés ? Le tableau ne donne pas d’analyse de ces aspects.
Quel est le taux global de réussite expérimentale des conceptions, en incluant les candidats échoués, et comment les écarts structuraux se traduisent-ils en performances fonctionnelles quantitatives ? Les résultats fournis ne permettent pas de répondre.