Abstract
Expression profiles of cancer are generally composed of three dimensions including gene probes, patients (e.g., metastasis or non-metastasis) and tissues (i.e., cancer or normal cells of a patient). In order to combine these three dimensions, we proposed a joint covariate detection that not only considered projections on gene probes and tissues simultaneously, but also concentrated on distinguishing patients into different groups. Due to highly lethal malignancy of hepatocellular carcinoma, we chose data GSE6857 to testify the effectiveness of our method. A bootstrap and accumulation strategy was introduced in, which could select candidate microRNAs to distinguish metastasis from non-metastasis patient group. Two pairs of microRNAs were further selected. Each component of either significant microRNA pair was derived from different cliques. Targets were sought and pathway analysis were made, which might reveal the mechanism of venous metastasis in primary hepatocellular carcinoma.
Highlights
These obtained statistical significances are faced with three major problems
We implemented the joint covariate detection approach on miRNA expression profiles of primary HCCs publicly available at the gene expression ominbus (GEO) with its accession number GSE685721, and extracted two miRNA pairs that might be associated with HCC venous metastasis
On assumption that expression profiles are composed of three dimensions including feature, sample and tissue, joint covariate detection embodies a bilinear projection on tissue and feature dimension, and integrative hypothesis testing (IHT) on sample dimension
Summary
These obtained statistical significances are faced with three major problems. A certain combination of cancer and adjacent normal expressions is to be made so that a better discriminative performance of selected features between two groups can be justified. Most of feature selection methods were based on hypothesis testing, which aimed to evaluate whether two populations of samples were significantly different or not by a certain discriminative statistics. On the basis of these insights, we proposed a joint covariate detection method that combined cancer and adjacent normal expression profiles and hypothesis testing and classification methods. As to each feature pair, expressions from cancer and adjacent normal tissues were viewed as a matrix form. The expressions of each sample on each feature pair formed a second-order matrix, which is composed of two column vectors derived from the the cancer and normal tissue of each feature. Potential target genes of these miRNAs were selected using TarBase[28] and the corresponding KEGG pathway was selected using DAVID29, which testified the significance of the selected miRNAs
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have