Network-based protein structural classification

Abstract

Experimental determination of protein function is resource-consuming. As an alternative, computational prediction of protein function has received attention. In this context, protein structural classification (PSC) can help, by allowing for determining structural classes of currently unclassified proteins based on their features, and then relying on the fact that proteins with similar structures have similar functions. Existing PSC approaches rely on sequence-based or direct three-dimensional (3D) structure-based protein features. By contrast, we first model 3D structures of proteins as protein structure networks (PSNs). Then, we use network-based features for PSC. We propose the use of graphlets, state-of-the-art features in many research areas of network science, in the task of PSC. Moreover, because graphlets can deal only with unweighted PSNs, and because accounting for edge weights when constructing PSNs could improve PSC accuracy, we also propose a deep learning framework that automatically learns network features from weighted PSNs. When evaluated on a large set of approximately 9400 CATH and approximately 12 800 SCOP protein domains (spanning 36 PSN sets), the best of our proposed approaches are superior to existing PSC approaches in terms of accuracy, with comparable running times. Our data and code are available at https://doi.org/10.5281/zenodo.3787922

Transfer Learning for Protein Structure Classification at Low Resolution

“…PDMs may be generated from the distances between different selections of atoms, including the alpha (C α ) and/or beta (C β ) carbons of the polypeptide backbone, or the heavy (non-hydrogen) atoms of the backbone and side-chains. The relative merits of different representations remain disputed: Duarte et al (2010) concluded that a combination of C α and C β atoms outperforms individual components (and particularly C α ) when reconstructing 3D protein structures from contact maps, whilst C α maps performed better than sidechain geometric centres for enzyme class prediction in a study by Da Silveira et al (2009), and heavy atom representations performed well in a more recent publication from Newaz et al (2020).…”

“…A review of the literature was conducted to identify historic approaches to PSC, detailed in Appendix Table A1. Three broad methodologies were encountered: those in which traditional machine learning algorithms were applied to features extracted from PDMs (Shi and Zhang, 2009;Taewijit and Waiyamai, 2010;Vani and Kumar, 2016;Pires et al, 2011); a second set training deep CNNs directly on large datasets of maps (Sikosek, 2019;Eguchi and Huang, 2020); and ensemble models combining different approaches (Zacharaki, 2017;Newaz et al, 2020). Studies relying on features derived from amino acid sequence alone are not listed exhaustively, however state of the art is included in Table A1 for completeness (Xia et al, 2017;Hou et al, 2018).…”

“…Among the best results in CATH classification have been those achieved using deep CNNs (Sikosek, 2019;Eguchi and Huang, 2020) and ensembles (Newaz et al, 2020). A modified version of DenseNet121, capable of simultaneous multi-class, multi-label prediction of CATH categories, demonstrated up to 87% accuracy on the most challenging task, being prediction of homologous superfamily from over 2000 possible classes (Sikosek, 2019).…”

“…It is important to note that the primary aim of the study was not accurate structure-level classification, but rather labelling individual amino acids according to CATH architecture, achieving an impressive average accuracy of 91%. Newaz et al (2020) combined models trained on different representations into ensembles for PSC. Among these, distances between the heavy atoms in a protein structure were described as ordered sub-graphs (graphlets), whose frequencies then served as an input feature for logistic regression.…”

Characterizing the organizational diversity of protein interaction networks across three domains of life

“…It has generally been argued that the topology of complex networks represent their underlying generative phenomena and there exist inherent similarity along with unique characteristic features among networks of different domains that can be leveraged to discriminate between them (12,13). Characterizing networks using these features have been reported for both within domain (category) (14) or different domain networks (15) using microscopic (e.g. degree), mesoscopic (e.g.…”