Revealing cancer subtypes with higher-order correlations applied to imaging and omics data

• DOI: 10.1186/s12920-017-0256-3
• Publication Date: 2017-03-31T01:53:08Z
• Authors: Kiley Graim; Tiffany Ting Liu; Achal S. Achrol; Evan O. Paull; Yulia Newton; Steven D. Chang; Griffith R. Harsh; Sergio P. Cordero; Daniel L. Rubin; Joshua M. Stuart
• Keywords: Human genetics; Radiogenomics

Patient stratification to identify subtypes with different disease manifestations, severity, and expected survival time is a critical task in cancer diagnosis treatment. While approaches using various biomarkers (including high-throughput gene expression measurements) for patient-to-patient comparisons have been successful elucidating previously unseen subtypes, there remains an untapped potential of incorporating genotypic phenotypic data discover novel or improved groupings. Here, we present HOCUS, unified analytical framework patient that uses community detection technique extract out sparse measurements. HOCUS constructs network from similarities the iteratively groups reconstructs into higher order clusters. We investigate merits higher-order correlations cluster samples patients terms their associations outcomes. In initial test method, approach identifies mutation glioblastoma, ovarian, breast, prostate, bladder cancers. several cases, provides improvement over molecular features directly compare samples. Application glioblastoma images reveals size location classification tumors improves human expert-based stratification. Subtypes based on can reveal comparable distinct The solutions provide biologically- treatment-relevant are just as significant original data.