2024 DeNovoDesignofHighAffinityProte
(Redirected from Zambaldi et al., 2024)
Jump to navigation
Jump to search
- (Zambaldi et al., 2024) ⇒ Vinicius Zambaldi, David La, Alexander E. Chu, Harshnira Patani, Amy E. Danson, Tristan O. C. Kwan, Thomas Frerix, Rosalia G. Schneider, David Saxton, Ashok Thillaisundaram, Zachary Wu, Isabel Moraes, Oskar Lange, Eliseo Papa, Gabriella Stanton, Victor Martin, Sukhdeep Singh, Lai H. Wong, Russ Bates, Simon A. Kohl, Josh Abramson, Andrew W. Senior, Yilmaz Alguel, Mary Y. Wu, Irene M. Aspalter, Katie Bentley, David L. V. Bauer, Peter Cherepanov, Demis Hassabis, Pushmeet Kohli, Rob Fergus, and Jue Wang. (2024). “De Novo Design of High-affinity Protein Binders with AlphaProteo.”
Subject Headings: Protein-Binding Protein, AlphaProteo, Protein Design.
Notes
- The paper introduces AlphaProteo, a machine learning model specifically designed for the de novo creation of protein binders, utilizing a novel deep learning architecture to achieve superior performance in generating high-affinity binders for target proteins.
- The paper demonstrates that AlphaProteo can produce binders with 3- to 300-fold better binding affinities compared to existing methods. This advance is achieved using a specialized deep generative model architecture that learns protein structure-function relationships and minimizes the need for multiple rounds of experimental testing.
- The paper showcases successful experimental validation of binders for seven structurally diverse proteins. The AI-driven system achieved binding success rates of 9% to 88%, far surpassing previous approaches by efficiently narrowing down candidate binders for in vitro testing.
- The paper reports the first computationally designed binders for specific targets, including challenging proteins like VEGF-A, a key protein target linked to cancer and diabetes. By applying learned patterns from a vast protein structure dataset, the model efficiently tackles previously difficult targets.
- The paper highlights AlphaProteo’s generalization capability. Using a multimodal neural network, AlphaProteo performs comparably across hundreds of targets from the PDB, illustrating its flexibility in binding to different structural and biochemical properties of proteins.
- The paper confirms the biological functionality of designed binders, demonstrating through experiments that binders can inhibit VEGF signaling in human cells and neutralize SARS-CoV-2 in monkey cells. This biological validation showcases the robustness of the designs generated by AlphaProteo.
- The paper validates the designed binder-target complex structures through Cryo-EM and X-ray crystallography, demonstrating that AlphaProteo’s predictions for protein-protein interactions align closely with experimentally observed molecular structures.
- The paper presents a novel generative model trained on protein structure and protein sequence data from the PDB and AlphaFold predictions, leveraging the AlphaFold-Multimer architecture to learn complex protein interaction patterns and predict accurate binder conformations with high specificity.
- The paper shows that AlphaProteo's machine learning model integrates a variational autoencoder and graph neural network (GNN) to generate binders ready for experimental use. This reduces the dependency on traditional high-throughput screening or manual affinity optimization, as the model captures relevant structural and chemical properties during generation.
- The paper emphasizes the use of AlphaFold as a key component in improving the success of designed binders. By integrating AlphaFold's prediction data into its architecture, AlphaProteo learns from a variety of structural protein features, enhancing its capacity to generalize across various biological targets.
- The paper highlights the role of an automated filtering process built into AlphaProteo's pipeline, which uses Bayesian optimization techniques to select only the most promising candidates for experimental validation. This automated filtering process improves the overall efficiency and accuracy of the in silico results, minimizing false positives during testing.
Cited By
Quotes
Abstract
Computational design of protein-binding proteins is a fundamental capability with broad utility in biomedical research and biotechnology. Recent methods have made strides against some target proteins, but on-demand creation of high-affinity binders without multiple rounds of experimental testing remains an unsolved challenge. This technical report introduces AlphaProteo, a family of machine learning models for protein design, and details its performance on the de novo binder design problem. With AlphaProteo, we achieve 3- to 300-fold better binding affinities and higher experimental success rates than the best existing methods on seven target proteins. Our results suggest that AlphaProteo can generate binders "ready-to-use" for many research applications using only one round of medium-throughput screening and no further optimization.
Experimental highlights:
- We introduce the AlphaProteo protein design system and experimentally test binders designed against eight structurally diverse target proteins.
- For seven of the targets, between 9% and 88% of the designs tested in the wet lab were experimentally verified as successful binders. These figures are higher than the best existing method and 5- to 100-fold higher than other methods. For one of these targets, we report the first computationally designed binders.
- The in silico performance of AlphaProteo on hundreds of target proteins from the PDB is comparable to these seven targets, suggesting that the method can potentially generalize widely. We chose one of the most challenging targets from this PDB screen as an 8th target but failed to obtain binders.
- We obtain binders with 80-960 picomolar affinities to four targets and low-nanomolar affinities to another three without needing high-throughput screening or experimental affinity optimization. For the seven targets, our designs have 3- to 300-fold better binding affinities than the best previously designed binder.
- We test binders for two of our targets for biological function, demonstrating inhibition of VEGF signaling in human cells and SARS-CoV-2 neutralization in Vero monkey cells.
- Cryo-EM and X-ray crystallography confirm the designed binder and binder-target complex structures.
References
;