K-nearest uphill clustering in the protein structure space

Handle URI:
http://hdl.handle.net/10754/622304
Title:
K-nearest uphill clustering in the protein structure space
Authors:
Cui, Xuefeng; Gao, Xin ( 0000-0002-7108-3574 )
Abstract:
The protein structure classification problem, which is to assign a protein structure to a cluster of similar proteins, is one of the most fundamental problems in the construction and application of the protein structure space. Early manually curated protein structure classifications (e.g., SCOP and CATH) are very successful, but recently suffer the slow updating problem because of the increased throughput of newly solved protein structures. Thus, fully automatic methods to cluster proteins in the protein structure space have been designed and developed. In this study, we observed that the SCOP superfamilies are highly consistent with clustering trees representing hierarchical clustering procedures, but the tree cutting is very challenging and becomes the bottleneck of clustering accuracy. To overcome this challenge, we proposed a novel density-based K-nearest uphill clustering method that effectively eliminates noisy pairwise protein structure similarities and identifies density peaks as cluster centers. Specifically, the density peaks are identified based on K-nearest uphills (i.e., proteins with higher densities) and K-nearest neighbors. To our knowledge, this is the first attempt to apply and develop density-based clustering methods in the protein structure space. Our results show that our density-based clustering method outperforms the state-of-the-art clustering methods previously applied to the problem. Moreover, we observed that computational methods and human experts could produce highly similar clusters at high precision values, while computational methods also suggest to split some large superfamilies into smaller clusters. © 2016 Elsevier B.V.
KAUST Department:
Computational Bioscience Research Center (CBRC); Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division
Citation:
Cui X, Gao X (2017) K-nearest uphill clustering in the protein structure space. Neurocomputing 220: 52–59. Available: http://dx.doi.org/10.1016/j.neucom.2016.04.065.
Publisher:
Elsevier BV
Journal:
Neurocomputing
KAUST Grant Number:
URF/1/1976-04
Issue Date:
26-Aug-2016
DOI:
10.1016/j.neucom.2016.04.065
Type:
Article
ISSN:
0925-2312
Sponsors:
This publication is based upon work supported by the King Abdullah University of Science and Technology (KAUST) Office of Sponsored Research (OSR) under Award No URF/1/1976-04. This research made use of the resources of the computer clusters at King Abdullah University of Science and Technology (KAUST).
Additional Links:
http://www.sciencedirect.com/science/article/pii/S0925231216309080
Appears in Collections:
Articles; Computational Bioscience Research Center (CBRC); Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division

Full metadata record

DC FieldValue Language
dc.contributor.authorCui, Xuefengen
dc.contributor.authorGao, Xinen
dc.date.accessioned2017-01-02T09:08:25Z-
dc.date.available2017-01-02T09:08:25Z-
dc.date.issued2016-08-26en
dc.identifier.citationCui X, Gao X (2017) K-nearest uphill clustering in the protein structure space. Neurocomputing 220: 52–59. Available: http://dx.doi.org/10.1016/j.neucom.2016.04.065.en
dc.identifier.issn0925-2312en
dc.identifier.doi10.1016/j.neucom.2016.04.065en
dc.identifier.urihttp://hdl.handle.net/10754/622304-
dc.description.abstractThe protein structure classification problem, which is to assign a protein structure to a cluster of similar proteins, is one of the most fundamental problems in the construction and application of the protein structure space. Early manually curated protein structure classifications (e.g., SCOP and CATH) are very successful, but recently suffer the slow updating problem because of the increased throughput of newly solved protein structures. Thus, fully automatic methods to cluster proteins in the protein structure space have been designed and developed. In this study, we observed that the SCOP superfamilies are highly consistent with clustering trees representing hierarchical clustering procedures, but the tree cutting is very challenging and becomes the bottleneck of clustering accuracy. To overcome this challenge, we proposed a novel density-based K-nearest uphill clustering method that effectively eliminates noisy pairwise protein structure similarities and identifies density peaks as cluster centers. Specifically, the density peaks are identified based on K-nearest uphills (i.e., proteins with higher densities) and K-nearest neighbors. To our knowledge, this is the first attempt to apply and develop density-based clustering methods in the protein structure space. Our results show that our density-based clustering method outperforms the state-of-the-art clustering methods previously applied to the problem. Moreover, we observed that computational methods and human experts could produce highly similar clusters at high precision values, while computational methods also suggest to split some large superfamilies into smaller clusters. © 2016 Elsevier B.V.en
dc.description.sponsorshipThis publication is based upon work supported by the King Abdullah University of Science and Technology (KAUST) Office of Sponsored Research (OSR) under Award No URF/1/1976-04. This research made use of the resources of the computer clusters at King Abdullah University of Science and Technology (KAUST).en
dc.publisherElsevier BVen
dc.relation.urlhttp://www.sciencedirect.com/science/article/pii/S0925231216309080en
dc.titleK-nearest uphill clustering in the protein structure spaceen
dc.typeArticleen
dc.contributor.departmentComputational Bioscience Research Center (CBRC)en
dc.contributor.departmentComputer, Electrical and Mathematical Sciences and Engineering (CEMSE) Divisionen
dc.identifier.journalNeurocomputingen
kaust.authorCui, Xuefengen
kaust.authorGao, Xinen
kaust.grant.numberURF/1/1976-04en
All Items in KAUST are protected by copyright, with all rights reserved, unless otherwise indicated.