Change point detection for clustered expression data.

BACKGROUND: To detect changes in biological processes, samples are often studied at several time points. We examined expression data measured at different developmental stages, or more broadly, historical data. Hence, the main assumption of our proposed methodology was the independence between the examined samples over time. In addition, however, the examinations were clustered at each time point by measuring littermates from relatively few mother mice at each developmental stage. As each examination was lethal, we had an independent data structure over the entire history, but a dependent data structure at a particular time point. Over the course of these historical data, we wanted to identify abrupt changes in the parameter of interest - change points. RESULTS: In this study, we demonstrated the application of generalized hypothesis testing using a linear mixed effects model as a possible method to detect change points. The coefficients from the linear mixed model were used in multiple contrast tests and the effect estimates were visualized with their respective simultaneous confidence intervals. The latter were used to determine the change point(s). In small simulation studies, we modelled different courses with abrupt changes and compared the influence of different contrast matrices. We found two contrasts, both capable of answering different research questions in change point detection: The Sequen contrast to detect individual change points and the McDermott contrast to find change points due to overall progression. We provide the R code for direct use with provided examples. The applicability of those tests for real experimental data was shown with in-vivo data from a preclinical study. CONCLUSION: Simultaneous confidence intervals estimated by multiple contrast tests using the model fit from a linear mixed model were capable to determine change points in clustered expression data. The confidence intervals directly delivered interpretable effect estimates representing the strength of the potential change point. Hence, scientists can define biologically relevant threshold of effect strength depending on their research question. We found two rarely used contrasts best fitted for detection of a possible change point: the Sequen and McDermott contrasts.

Heterogeneous Effects of Direct Hypoxia Pathway Activation in Kidney Cancer.

General activation of hypoxia-inducible factor (HIF) pathways is classically associated with adverse prognosis in cancer and has been proposed to contribute to oncogenic drive. In clear cell renal carcinoma (CCRC) HIF pathways are upregulated by inactivation of the von-Hippel-Lindau tumor suppressor. However HIF-1α and HIF-2α have contrasting effects on experimental tumor progression. To better understand this paradox we examined pan-genomic patterns of HIF DNA binding and associated gene expression in response to manipulation of HIF-1α and HIF-2α and related the findings to CCRC prognosis. Our findings reveal distinct pan-genomic organization of canonical and non-canonical HIF isoform-specific DNA binding at thousands of sites. Overall associations were observed between HIF-1α-specific binding, and genes associated with favorable prognosis and between HIF-2α-specific binding and adverse prognosis. However within each isoform-specific set, individual gene associations were heterogeneous in sign and magnitude, suggesting that activation of each HIF-α isoform contributes a highly complex mix of pro- and anti-tumorigenic effects.