Graph-based semi-supervised learning with genomic data integration using condition-responsive genes applied to phenotype classification. (13th May 2017)

• Record Type: Journal Article
• Authors: Doostparast Torshizi, Abolfazl; Petzold, Linda R
• Is Part Of: Journal of the American Medical Informatics Association. Volume 25:Number 1(2018)
• Page Start: 99
• Page End: 108
• Publication Date: 2017-05-13
• Subjects: gene expression -- DNA methylation -- semi-supervised learning -- graph theory -- ovarian cancer -- data integration; Medical informatics -- Periodicals; Information Services -- Periodicals; Medical Informatics -- Periodicals; Médecine -- Informatique -- Périodiques; Informatica; Geneeskunde; Informatique médicale; Computer network resources; Electronic journals; 610.285
• DOI: 10.1093/jamia/ocx032
• Languages: English
• ISSNs: 1067-5027

Abstract

Abstract: Objective: Data integration methods that combine data from different molecular levels such as genome, epigenome, transcriptome, etc., have received a great deal of interest in the past few years. It has been demonstrated that the synergistic effects of different biological data types can boost learning capabilities and lead to a better understanding of the underlying interactions among molecular levels. Methods: In this paper we present a graph-based semi-supervised classification algorithm that incorporates latent biological knowledge in the form of biological pathways with gene expression and DNA methylation data. The process of graph construction from biological pathways is based on detecting condition-responsive genes, where 3 sets of genes are finally extracted: all condition responsive genes, high-frequency condition-responsive genes, and P -value–filtered genes. Results: The proposed approach is applied to ovarian cancer data downloaded from the Human Genome Atlas. Extensive numerical experiments demonstrate superior performance of the proposed approach compared to other state-of-the-art algorithms, including the latest graph-based classification techniques. Conclusions: Simulation results demonstrate that integrating various data types enhances classification performance and leads to a better understanding of interrelations between diverse omics data types. The proposed approach outperforms many of the state-of-the-art data integration algorithms.