Bioinformatics approaches to heterogeneous omic data integration.

摘要

With the advent of whole-genome sequencing and high-throughput microarray experimental technologies, an important challenge confronting researchers lies in analyzing these large-scale data sets and extracting discernable biological information from them. Existing statistical or bioinformatical approaches have known limitations on extracting the most compact sets of genes to distinguish between tumor subtype/phenotypes and integrating heterogeneous high dimensional omic data effectively. Therefore, powerful approaches/pipelines that would facilitate data management, visualization and integration for biological findings are warranted, and should be executed without sacrificing statistical rigor and computational efficiency. Multivariate analyses that target characterization of the complex and hierarchical interplay between multiple genes at various levels would be useful as well. To overcome known limitations of standard statistical approaches, two specific projects were proposed. First, with the goal to improve the prediction accuracy while considering sparsity, a non-parametric, iterative algorithm, Splitting Random Forest (SRF), was developed, to robustly identify genes that distinguish between subtypes/phenotypes. The manuscript from this new algorithm is now published at Journal of Clinical Bioinformatics. Second, I focus on integration of the omic data sequentially using currently acceptable bioinformatical approaches to derive biological insight for glioma progression by applying known glioblastoma (GBM) expression and methylation based subtypes to lower grade gliomas (LGGs; grade II-III gliomas), while examining copy number changes. The manuscript from this analysis is now under review at a top tier cancer journal.