Quantitative protein descriptors for secondary structure characterization and protein classification
2009 (English)In: Chemometrics and Intelligent Laboratory Systems, ISSN 0169-7439, E-ISSN 1873-3239, Vol. 95, no 1, 74-85 p.Article in journal (Refereed) Published
In this study protein chains were characterized based on alignment-independent protein descriptors using three types of structural and sequence data; (i) C-α atom Euclidean distances, (ii) protein backbone ψ and φ angles and (iii) amino acid physicochemical properties (zz-scales). The descriptors were analyzed using principal component analysis (PCA) and further elucidated using the multivariate methods partial least-squares projections to latent structures discriminant-analysis (PLS-DA) and hierarchical-PLS-DA. The descriptors were applied to three protein chain datasets: (i) 82 chains classified, according to the structural classification of proteins (SCOP) scheme, as either all-α or all-β; (ii) 96 chains classified as either α + β or α/β and (iii) 6590 chains of all aforementioned classes selected from the PDB-select database. Results showed that the descriptors related to the secondary structure of the chains. The C-α Euclidean distances, and as expected, the protein backbone angles were found to be most important for the characterization and classification of chains. Assignment of SCOP classes using PLS-DA based on all descriptor types was satisfactory for all-α and all-β chains with more than 93% correct classifications of a large external test set, while the protein chains of types α/β and α + β was harder to discriminate between, resulting in 74% and 54% correct classifications, respectively.
Place, publisher, year, edition, pages
2009. Vol. 95, no 1, 74-85 p.
Multivariate analysis, Protein descriptor, SCOP, Auto covariance, Auto cross-covariance
IdentifiersURN: urn:nbn:se:umu:diva-3651DOI: 10.1016/j.chemolab.2008.08.006OAI: oai:DiVA.org:umu-3651DiVA: diva2:142448