Evolutionary-scale prediction of atomic-level protein structure with a language model

Science, Journal Year: 2023, Volume and Issue: 379(6637), P. 1123 - 1130

Published: March 16, 2023

Recent advances in machine learning have leveraged evolutionary information multiple sequence alignments to predict protein structure. We demonstrate direct inference of full atomic-level structure from primary using a large language model. As models sequences are scaled up 15 billion parameters, an atomic-resolution picture emerges the learned representations. This results order-of-magnitude acceleration high-resolution prediction, which enables large-scale structural characterization metagenomic proteins. apply this capability construct ESM Metagenomic Atlas by predicting structures for >617 million sequences, including >225 that predicted with high confidence, gives view into vast breadth and diversity natural

Language: Английский

---

UCSF ChimeraX: Tools for structure building and analysis

Protein Science, Journal Year: 2023, Volume and Issue: 32(11)

Published: Sept. 29, 2023

Advances in computational tools for atomic model building are leading to accurate models of large molecular assemblies seen electron microscopy, often at challenging resolutions 3-4 Å. We describe new methods the UCSF ChimeraX modeling package that take advantage machine-learning structure predictions, provide likelihood-based fitting maps, and compute per-residue scores identify errors. Additional model-building assist analysis mutations, post-translational modifications, interactions with ligands. present latest capabilities, including several community-developed extensions. is available free charge noncommercial use https://www.rbvi.ucsf.edu/chimerax.

Language: Английский

---

Harnessing protein folding neural networks for peptide–protein docking

Nature Communications, Journal Year: 2022, Volume and Issue: 13(1)

Published: Jan. 10, 2022

Highly accurate protein structure predictions by deep neural networks such as AlphaFold2 and RoseTTAFold have tremendous impact on structural biology beyond. Here, we show that, although these learning approaches originally been developed for the in silico folding of monomers, also enables quick modeling peptide-protein interactions. Our simple implementation generates complex models without requiring multiple sequence alignment information peptide partner, can handle binding-induced conformational changes receptor. We explore what has memorized learned, describe specific examples that highlight differences compared to state-of-the-art docking protocol PIPER-FlexPepDock. These results holds great promise providing insight into a wide range complexes, serving starting point detailed characterization manipulation

Language: Английский

---

Scientific discovery in the age of artificial intelligence

Nature, Journal Year: 2023, Volume and Issue: 620(7972), P. 47 - 60

Published: Aug. 2, 2023

Language: Английский

---

Large language models generate functional protein sequences across diverse families

Nature Biotechnology, Journal Year: 2023, Volume and Issue: 41(8), P. 1099 - 1106

Published: Jan. 26, 2023

Language: Английский

---

A structural biology community assessment of AlphaFold2 applications

Nature Structural & Molecular Biology, Journal Year: 2022, Volume and Issue: 29(11), P. 1056 - 1067

Published: Nov. 1, 2022

Most proteins fold into 3D structures that determine how they function and orchestrate the biological processes of cell. Recent developments in computational methods for protein structure predictions have reached accuracy experimentally determined models. Although this has been independently verified, implementation these across structural-biology applications remains to be tested. Here, we evaluate use AlphaFold2 (AF2) study characteristic structural elements; impact missense variants; ligand binding site predictions; modeling interactions; experimental data. For 11 proteomes, an average 25% additional residues can confidently modeled when compared with homology modeling, identifying features rarely seen Protein Data Bank. AF2-based disorder complexes surpass dedicated tools, AF2 models used diverse equally well structures, confidence metrics are critically considered. In summary, find advances likely a transformative biology broader life-science research.

Language: Английский

---

ProtGPT2 is a deep unsupervised language model for protein design

Nature Communications, Journal Year: 2022, Volume and Issue: 13(1)

Published: July 27, 2022

Protein design aims to build novel proteins customized for specific purposes, thereby holding the potential tackle many environmental and biomedical problems. Recent progress in Transformer-based architectures has enabled implementation of language models capable generating text with human-like capabilities. Here, motivated by this success, we describe ProtGPT2, a model trained on protein space that generates de novo sequences following principles natural ones. The generated display amino acid propensities, while disorder predictions indicate 88% ProtGPT2-generated are globular, line sequences. Sensitive sequence searches databases show ProtGPT2 distantly related ones, similarity networks further demonstrate is sampling unexplored regions space. AlphaFold prediction ProtGPT2-sequences yields well-folded non-idealized structures embodiments large loops reveals topologies not captured current structure databases. matter seconds freely available.

Language: Английский

---

Sampling alternative conformational states of transporters and receptors with AlphaFold2

eLife, Journal Year: 2022, Volume and Issue: 11

Published: March 3, 2022

Equilibrium fluctuations and triggered conformational changes often underlie the functional cycles of membrane proteins. For example, transporters mediate passage molecules across cell membranes by alternating between inward- outward-facing states, while receptors undergo intracellular structural rearrangements that initiate signaling cascades. Although plasticity these proteins has historically posed a challenge for traditional de novo protein structure prediction pipelines, recent success AlphaFold2 (AF2) in CASP14 culminated modeling transporter multiple conformations to high accuracy. Given AF2 was designed predict static structures proteins, it remains unclear if this result represents an underexplored capability accurately and/or heterogeneity. Here, we present approach drive sample alternative topologically diverse G-protein-coupled are absent from training set. Whereas models most generated using default pipeline conformationally homogeneous nearly identical one another, reducing depth input sequence alignments stochastic subsampling led generation accurate conformations. In our benchmark, spanned range two experimental interest, with at extremes distributions observed be among (average template score 0.94). These results suggest straightforward identifying native-like also highlighting need next deep learning algorithms ensembles biophysically relevant states.

Language: Английский

---

High-resolutionde novostructure prediction from primary sequence

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2022, Volume and Issue: unknown

Published: July 22, 2022

Abstract Recent breakthroughs have used deep learning to exploit evolutionary information in multiple sequence alignments (MSAs) accurately predict protein structures. However, MSAs of homologous proteins are not always available, such as with orphan or fast-evolving like antibodies, and a typically folds natural setting from its primary amino acid into three-dimensional structure, suggesting that should be necessary protein’s folded form. Here, we introduce OmegaFold, the first computational method successfully high-resolution structure single alone. Using new combination language model allows us make predictions sequences geometry-inspired transformer trained on structures, OmegaFold outperforms RoseTTAFold achieves similar prediction accuracy AlphaFold2 recently released enables accurate do belong any functionally characterized family antibodies tend noisy due fast evolution. Our study fills much-encountered gap brings step closer understanding folding nature.

Language: Английский

---

Evolutionary-scale prediction of atomic level protein structure with a language model

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2022, Volume and Issue: unknown

Published: July 21, 2022

Abstract Artificial intelligence has the potential to open insight into structure of proteins at scale evolution. It only recently been possible extend protein prediction two hundred million cataloged proteins. Characterizing structures exponentially growing billions sequences revealed by large gene sequencing experiments would necessitate a break-through in speed folding. Here we show that direct inference from primary sequence using language model enables an order magnitude speed-up high resolution prediction. Leveraging models learn evolutionary patterns across millions sequences, train up 15B parameters, largest date. As are scaled they information three-dimensional individual atoms. This results is 60x faster than state-of-the-art while maintaining and accuracy. Building on this, present ESM Metage-nomic Atlas. first large-scale structural characterization metagenomic proteins, with more 617 structures. The atlas reveals 225 confidence predictions, including whose novel comparison experimentally determined structures, giving unprecedented view vast breadth diversity some least understood earth.

Language: Английский

---