An Integrated Robust Graph Regularized Non-negative Matrix Factorization for Multi-dimensional Genomic Data Analysis

Abstract

With the generation of “multi-dimensional genomic data”, the multi-platform genomic analysis technology for simultaneous biological samples has been rapidly developed. However, the existing shortcomings of the models for comprehensive analysis of multi-dimensional genomic data are the lack of robustness. We use the integrated matrix factorization model by introducing L2,1-norm to deal with above problem. In this paper, we propose an Integrated Robust Graph Regularization Non-negative Matrix Factorization for multi-dimensional genomic data analysis which named iRGNMF. In the formulation of the objective function, we introduced the local structure to maintain regularization and L2,1-norm to consider the data geometry and model robustness. We also applied this model to three types of data of the same cancer from The Cancer Genome Atlas (TCGA). Experiments shown that iRGNMF has obtained considerable effects in sample clustering, and found suspicious disease genes, which are used to reveal hidden patterns and information of biology in multi-dimensional gene data.