Clustering predicted structures at the scale of the known protein universe

Abstract

Proteins are key to all cellular processes and their structure is important in understanding their function and evolution. Sequence-based predictions of protein structures have increased in accuracy with over 214 million predicted structures available in the AlphaFold database (AFDB). However, studying protein structures at this scale requires highly efficient methods. Here, we developed a structural-alignment based clustering algorithm - Foldseek cluster - that can cluster hundreds of millions of structures. Using this method we have clustered all structures in AFDB, identifying 2.27M non-singleton structural clusters, of which 31% lack annotations representing likely novel structures. Clusters without annotation tend to have few representatives covering only 4% of all proteins in the AFDB. Evolutionary analysis suggests that most clusters are ancient in origin but 4% seem species specific, representing lower quality predictions or examples of de-novo gene birth. Additionally, we show how structural comparisons can be used to predict domain families and their relationships, identifying examples of remote homology. Based on these analyses we identify several examples of human immune related proteins with remote homology in prokaryotic species which illustrates the value of this resource for studying protein function and evolution across the tree of life.

What is hidden in the darkness? Deep-learning assisted large-scale protein family curation uncovers novel protein families and folds

“…The availability of high quality predicted structures allows us to expand functional curation efforts to include structure-based similarities. One useful approach for this is structure-based alignment, which is described in (Barrio-Hernandez et al, 2023), and also incorporated into our examples to find similarities to proteins with known PDB structures using Foldseek. Here, we look into another angle of structure comparison, which is based on the concept of a “structural outlier”.…”

“…Recently, several complementary approaches have been developed to categorise the diversity of the protein universe and uncover novelties (Akdel et al, 2022; Barrio-Hernandez et al, 2023; Bordin et al, 2023), again highlighting the importance of incorporating multiple perspectives and methods in protein function annotation. These approaches showcase the significance of using a diverse set of information to gain a more complete understanding of protein function and its role in cellular processes.…”

Opportunities and Challenges for Machine Learning-Assisted Enzyme Engineering

“…Clustering similar structures is one way to annotate for function. 63 Alternatively, many enzymes have common "modules," or recurring residue arrangements, which perform similar reactions. 64 The structures of active sites in unlabeled protein structures could be compared to existing structures to identify new, diverse sets of proteins with given function, using models trained on sequence and structure.…”

Dual‐wield NTPases: A novel protein family mined from AlphaFold DB

“…Dedicated data mining demands a clearly stated working hypothesis. While several groups have pursued intensive model classifications against AlphaFold DB (Barrio‐Hernandez et al, 2023; Bordin et al, 2023; Durairaj et al, 2023), this bird's‐eye approach could miss unique and intriguing proteins. To find these hidden gems, we defined a very specific database search question: are there monomeric proteins that contain multiple phosphate‐binding loops (P‐loops) on a single continuous β‐sheet?…”