Highly accurate protein structure prediction for the human proteome

Tunyasuvunakool, Kathryn; Adler, Jonas; Wu, Zachary; Green, Tim; Zieliński, Michał; Žídek, Augustin; Bridgland, Alex; Cowie, Andrew; Meyer, Clemens; Laydon, Agata; Velankar, Sameer; Kleywegt, Gerard J.; Bateman, Alex; Evans, Rhett; Pritzel, Alexander; Figurnov, Michael; Ronneberger, Olaf; Bates, Russ; Kohl, Simon A. A.; Potapenko, Anna; Ballard, Andrew J.; Romera-Paredes, Bernardino; Nikolov, Stanislav; Jain, Rishub; Clancy, Ellen; Reiman, David; Petersen, Stig; Senior, Andrew W.; Kavukcuoglu, Koray; Birney, Ewan; Kohli, Pushmeet; Jumper, John; Hassabis, Demis

doi:10.1038/s41586-021-03828-1

Cited by 2,320 publications

(2,040 citation statements)

References 77 publications

Supporting

Mentioning

1,825

Contrasting

Unclassified

Order By: Relevance

“…2 determination (Flower & Hurley, 2021). However, even this pinnacle of 50 years of research has two major shortcomings: (i) predictions are more familyspecific than protein-specific, (ii) structure prediction requires substantial computing resources, although 3D structure predictions have been made available for 20 entire proteomes with more to come soon (Tunyasuvunakool et al, 2021).…”

Section: Introduction mentioning

confidence: 99%

Protein language model embeddings for fast, accurate, alignment-free protein structure prediction

Weißenow

Heinzinger

Rost

2021

Preprint

View full text Add to dashboard Cite

All state-of-the-art (SOTA) protein structure predictions rely on evolutionary information captured in multiple sequence alignments (MSAs), primarily on evolutionary couplings (co-evolution). Such information is not available for all proteins and is computationally expensive to generate. Prediction models based on Artificial Intelligence (AI) using only single sequences as input are easier and cheaper but perform so poorly that speed becomes irrelevant. Here, we described the first competitive AI solution exclusively inputting embeddings extracted from pre-trained protein Language Models (pLMs), namely from the transformer pLM ProtT5, from single sequences into a relatively shallow (few free parameters) convolutional neural network (CNN) trained on inter-residue distances, i.e. protein structure in 2D. The major advance originated from processing the attention heads learned by ProtT5. Although these models required at no point any MSA, they matched the performance of methods relying on co-evolution. Although not reaching the very top, our lean approach came close at substantially lower costs thereby speeding up development and each future prediction. By generating protein-specific rather than family-averaged predictions, these new solutions could distinguish between structural features differentiating members of the same family of proteins with similar structure predicted alike by all other top methods.

show abstract

Section: Introduction mentioning

confidence: 99%

Protein language model embeddings for fast, accurate, alignment-free protein structure prediction

Weißenow

Heinzinger

Rost

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Phenix Resolve Density Modification was applied to maps for visualization 52 . The N-terminus prior to TM1 was copied from AlphaFold2 predicted structures 53,54 and refined into the cryo-EM density. Structures were analyzed and figures were prepared with HOLE 55 , DALI 27 , PyMOL, ChimeraX 56 , JalView 57 , Prism 8, GNU Image Manipulation Program, and Adobe Photoshop and Illustrator software.…”

Section: Modeling Refinement and Analysismentioning

confidence: 99%

Structures of Tweety Homolog Proteins TTYH2 and TTYH3 reveal a Ca²⁺-dependent switch from intra- to inter-membrane dimerization

Hoel

Brohawn

2021

Preprint

View full text Add to dashboard Cite

Tweety homologs (TTYHs) comprise a conserved family of transmembrane proteins found in eukaryotes with three members (TTYH1-3) in vertebrates. They are widely expressed in mammals including at high levels in the nervous system and have been implicated in cancers and other diseases including epilepsy, chronic pain, and viral infections. TTYHs have been reported to form Ca2+- and cell volume-regulated anion channels structurally distinct from any characterized protein family with potential roles in cell adhesion, migration, and developmental signaling. To provide insight into TTYH family structure and function, we determined cryo-EM structures of Mus musculus TTYH2 and TTYH3 in lipid nanodiscs. TTYH2 and TTYH3 adopt a novel fold which includes an extended extracellular domain with a partially solvent exposed pocket that may be an interaction site for hydrophobic molecules. In the presence of Ca2+, TTYH2 and TTYH3 form homomeric cis-dimers bridged by extracellularly coordinated Ca2+. Strikingly, in the absence of Ca2+, TTYH2 forms trans-dimers that span opposing membranes across a ~130 Å intermembrane space as well as a monomeric state. All TTYH structures lack ion conducting pathways and we do not observe TTYH2-dependent channel activity in cells. We conclude TTYHs are not pore forming subunits of anion channels and their function may involve Ca2+-dependent changes in quaternary structure, interactions with hydrophobic molecules near the extracellular membrane surface, and/or association with additional protein partners.

show abstract

“…RoseTTAFold 8 was developed using a similar end-to-end network architecture and achieved structure prediction accuracy close to that of AlphaFold2. The high accuracy of AlphaFlold2 make it possible to offer meaningful predicted structures to biologists, which is applied to human genome 9 . In this work, we focus on the SLC fold atlas and attempt to provide more knowledge about the function of human SLC superfamily by integrating predicted structures from multiple different methods.…”

Section: Protocol To Identify New Foldsmentioning

confidence: 99%

Rational Exploration of Fold Atlas for Human Solute Carrier Proteins

Xie

Chi

Huang

et al. 2021

Preprint

View full text Add to dashboard Cite

Solute carrier superfamily (SLC) is the largest group responsible for transmembrane transport of substances in human cells. It includes more than 400 members that can be divided into 65 families according to their physiological function and sequence similarity. Different families of solute transporters can adopt the same or different folds that determines the working mechanism and reflects the evolutionary relationship between SLC members. Analysis of structural data shows that there are 13 different folds in the solute carrier superfamily covering 40 families and 342 members. To further study their working mechanism, we systematically explored the SLC superfamily to find more folds. Our results indicate that there are at least three new folds in the solute transporter superfamily among which the SLC44 family is experimentally verified to have a new fold. Our work has laid a foundation and provided important insights for the systematic and comprehensive study on the structure and function of solute carriers.

show abstract

Highly accurate protein structure prediction for the human proteome

Cited by 2,320 publications

References 77 publications

Protein language model embeddings for fast, accurate, alignment-free protein structure prediction

Protein language model embeddings for fast, accurate, alignment-free protein structure prediction

Structures of Tweety Homolog Proteins TTYH2 and TTYH3 reveal a Ca²⁺-dependent switch from intra- to inter-membrane dimerization

Rational Exploration of Fold Atlas for Human Solute Carrier Proteins

Contact Info

Product

Resources

About

Highly accurate protein structure prediction for the human proteome

Cited by 2,320 publications

References 77 publications

Protein language model embeddings for fast, accurate, alignment-free protein structure prediction

Protein language model embeddings for fast, accurate, alignment-free protein structure prediction

Structures of Tweety Homolog Proteins TTYH2 and TTYH3 reveal a Ca2+-dependent switch from intra- to inter-membrane dimerization

Rational Exploration of Fold Atlas for Human Solute Carrier Proteins

Contact Info

Product

Resources

About

Structures of Tweety Homolog Proteins TTYH2 and TTYH3 reveal a Ca²⁺-dependent switch from intra- to inter-membrane dimerization