Fig. 3: Topology provides organising principles for predicted protein structures.

Topological clusters in dimension 1 often provide a refinement of protein domains. As an example, we see protein kinase, coloured by its two CATH domains (A) and topological clusters (B). C Clustering homogeneity can be used to check if a topological cluster contains only residues belonging to a single domain, with 1 corresponding to a perfect sub-partition, and 0 to each cluster containing all the same labels. The bar plot shows the distribution of homogeneity scores for a set of 38,171 non-redundant AlphaFold2 predictions with identified CATH domains. D Two-dimensional topological cluster boundary points are enriched for binding sites from the Mechanism and Catalytic Site Atlas (M-CSA) dataset. The bar chart shows the distribution of distances (in number of residues) from cluster boundaries to residues of either binding sites or other residues. E The topological analysis is robust to small perturbations. The bar plot shows the distribution of correlation coefficients between the topological influence scores for AlphaFold2 predicted structures and their experimentally solved counterparts. The high values show that topological features tend to correspond and to interest the same residues. F Topological clusters for horse apomyoglobin (UniProt accession P68082) differentiated by colour. Folding events based on experimental evidence is indicated by transparency, where fully opaque sections are formed. Source data are provided as a Source Data file in Source_data/source_data_Figure3C-S8.csv, Source_data/source_data_Figure3D.csv, and Source_data/source_data_Figure3E.csv.