De novo design of protein interactions with learned surface fingerprints

Nature, Год журнала: 2023, Номер 617(7959), С. 176 - 184

Опубликована: Апрель 26, 2023

Abstract Physical interactions between proteins are essential for most biological processes governing life 1 . However, the molecular determinants of such have been challenging to understand, even as genomic, proteomic and structural data increase. This knowledge gap has a major obstacle comprehensive understanding cellular protein–protein interaction networks de novo design protein binders that crucial synthetic biology translational applications 2–9 Here we use geometric deep-learning framework operating on surfaces generates fingerprints describe chemical features critical drive 10 We hypothesized these capture key aspects recognition represent new paradigm in computational novel interactions. As proof principle, computationally designed several engage four targets: SARS-CoV-2 spike, PD-1, PD-L1 CTLA-4. Several designs were experimentally optimized, whereas others generated purely silico, reaching nanomolar affinity with mutational characterization showing highly accurate predictions. Overall, our surface-centric approach captures physical recognition, enabling an and, more broadly, artificial function.

Язык: Английский

---

Protein–protein interaction prediction with deep learning: A comprehensive review

Computational and Structural Biotechnology Journal, Год журнала: 2022, Номер 20, С. 5316 - 5341

Опубликована: Янв. 1, 2022

Most proteins perform their biological function by interacting with themselves or other molecules. Thus, one may obtain insights into protein functions, disease prevalence, and therapy development identifying protein–protein interactions (PPI). However, finding the non-interacting pairs through experimental approaches is labour-intensive time-consuming, owing to variety of proteins. Hence, interaction protein–ligand binding problems have drawn attention in fields bioinformatics computer-aided drug discovery. Deep learning methods paved way for scientists predict 3-D structure from genomes, functions attributes a protein, modify design new provide desired functions. This review focuses on recent deep applied including predicting sites, binding, design.

Язык: Английский

---

De novo protein design—From new structures to programmable functions

Cell, Год журнала: 2024, Номер 187(3), С. 526 - 544

Опубликована: Фев. 1, 2024

Methods from artificial intelligence (AI) trained on large datasets of sequences and structures can now "write" proteins with new shapes molecular functions de novo, without starting found in nature. In this Perspective, I will discuss the state field novo protein design at juncture physics-based modeling approaches AI. New folds higher-order assemblies be designed considerable experimental success rates, difficult problems requiring tunable control over conformations precise shape complementarity for recognition are coming into reach. Emerging incorporate engineering principles-tunability, controllability, modularity-into process beginning. Exciting frontiers lie deconstructing cellular and, conversely, constructing synthetic signaling ground up. As methods improve, many more challenges unsolved.

Язык: Английский

---

De novo protein design by inversion of the AlphaFold structure prediction network

Protein Science, Год журнала: 2023, Номер 32(6)

Опубликована: Май 11, 2023

De novo protein design enhances our understanding of the principles that govern folding and interactions, has potential to revolutionize biotechnology through engineering novel functionalities. Despite recent progress in computational strategies, de structures remains challenging, given vast size sequence-structure space. AlphaFold2 (AF2), a state-of-the-art neural network architecture, achieved remarkable accuracy predicting from amino acid sequences. This raises question whether AF2 learned sufficiently for design. Here, we sought answer this by inverting network, using prediction weight set loss function bias generated sequences adopt target fold. Initial trials resulted designs with an overrepresentation hydrophobic residues on surface compared their natural family, requiring additional optimization. In silico validation showed correct fold, hydrophilic densely packed core. vitro 7 out 39 were folded stable solution high melting temperatures. summary, workflow solely based does not seem fully capture basic design, as observed surface's vs. patterning. However, minimal post-design intervention, these pipelines viable assessed experimental characterization. Thus, such show contribute solving outstanding challenges

Язык: Английский

---

Opportunities and challenges in design and optimization of protein function

Nature Reviews Molecular Cell Biology, Год журнала: 2024, Номер 25(8), С. 639 - 653

Опубликована: Апрель 2, 2024

Язык: Английский

---

Targeting protein–ligand neosurfaces with a generalizable deep learning tool

Nature, Год журнала: 2025, Номер unknown

Опубликована: Янв. 15, 2025

Molecular recognition events between proteins drive biological processes in living systems1. However, higher levels of mechanistic regulation have emerged, which protein–protein interactions are conditioned to small molecules2–5. Despite recent advances, computational tools for the design new chemically induced protein remained a challenging task field6,7. Here we present strategy that target neosurfaces, is, surfaces arising from protein–ligand complexes. To develop this strategy, leveraged geometric deep learning approach based on learned molecular surface representations8,9 and experimentally validated binders against three drug-bound complexes: Bcl2–venetoclax, DB3–progesterone PDF1–actinonin. All demonstrated high affinities accurate specificities, as assessed by mutational structural characterization. Remarkably, fingerprints previously trained only could be applied neosurfaces with molecules, providing powerful demonstration generalizability is uncommon other approaches. We anticipate such designed will potential expand sensing repertoire assembly synthetic pathways engineered cells innovative drug-controlled cell-based therapies10. A used formed interactions, applications development therapeutic modalities glues or therapies.

Язык: Английский

---

Computational Methods in Immunology and Vaccinology: Design and Development of Antibodies and Immunogens

Journal of Chemical Theory and Computation, Год журнала: 2023, Номер 19(16), С. 5315 - 5333

Опубликована: Авг. 1, 2023

The design of new biomolecules able to harness immune mechanisms for the treatment diseases is a prime challenge computational and simulative approaches. For instance, in recent years, antibodies have emerged as an important class therapeutics against spectrum pathologies. In cancer, immune-inspired approaches are witnessing surge thanks better understanding tumor-associated antigens their engagement or evasion from human system. Here, we provide summary main state-of-the-art that used antigens, parallel, review key methodologies epitope identification both B- T-cell mediated responses. A special focus devoted description structure- physics-based models, privileged over purely sequence-based We discuss implications novel methods engineering with tailored immunological properties possible therapeutic uses. Finally, highlight extraordinary challenges opportunities presented by integration emerging Artificial Intelligence technologies prediction epitopes, antibodies.

Язык: Английский

---

Deploying synthetic coevolution and machine learning to engineer protein-protein interactions

Science, Год журнала: 2023, Номер 381(6656)

Опубликована: Июль 27, 2023

Fine-tuning of protein-protein interactions occurs naturally through coevolution, but this process is difficult to recapitulate in the laboratory. We describe a platform for synthetic coevolution that can isolate matched pairs interacting muteins from complex libraries. This large dataset coevolved complexes drove systems-level analysis molecular recognition between Z domain–affibody spanning wide range structures, affinities, cross-reactivities, and orthogonalities, captured broad spectrum coevolutionary networks. Furthermore, we harnessed pretrained protein language models expand, silico, amino acid diversity our screen, predicting remodeled interfaces beyond reach experimental library. The integration these approaches provides means simulating generating with diverse properties biotechnology biology.

Язык: Английский

---

Deep learning for advancing peptide drug development: Tools and methods in structure prediction and design

European Journal of Medicinal Chemistry, Год журнала: 2024, Номер 268, С. 116262 - 116262

Опубликована: Фев. 19, 2024

Язык: Английский

---

Integrating Dynamic Network Analysis with AI for Enhanced Epitope Prediction in PD-L1:Affibody Interactions

Journal of the American Chemical Society, Год журнала: 2024, Номер 146(34), С. 23842 - 23853

Опубликована: Авг. 15, 2024

Understanding binding epitopes involved in protein–protein interactions and accurately determining their structure are long-standing goals with broad applicability industry biomedicine. Although various experimental methods for epitope determination exist, these approaches typically low throughput cost-intensive. Computational have potential to accelerate predictions; however, recently developed artificial intelligence (AI)-based frequently fail predict of synthetic domains few natural homologues. Here we an integrated method employing generalized-correlation-based dynamic network analysis on multiple molecular dynamics (MD) trajectories, initiated from AlphaFold2Multimer structures, unravel the therapeutic PD-L1:Affibody complex. Both AlphaFold2 conventional trajectory were ineffective distinguishing between two proposed models, parallel perpendicular. However, our approach, utilizing analysis, demonstrated that perpendicular mode was significantly more stable. These predictions validated using a suite mapping protocols, including cross-linking mass spectrometry next-generation sequencing-based deep mutational scanning. Conversely, AlphaFold3 failed bound pose, highlighting necessity exploratory research search challenging notion AI-generated protein structures can be accepted without scrutiny. Our underscores enhance AI-based accurate identification interaction interfaces.

Язык: Английский

---