Biostatistics Core

The Biostatistics Core provides tools to enhance the rigor of all the CMCR studies. The Core will provide support for investigators in each Project by selecting and implementing appropriate statistical and mathematical modeling methodologies for each experimental design. These activities include data analysis and statistical testing. The Core will maintain constant interaction with the Projects, establishing a productive “cycle”, where data analysis results enhance our understanding of the biological effects of complex radiation exposures, and improve experimental design and hypothesis generation and testing. Specifically, the Core will implement techniques such as parametric regression with linear or generalized linear models, information theoretic multimodel inference, and ensemble machine learning with gradient boosting and random forests. To analyze multidimensional transcriptomics and metabolomics data sets, dimension reduction and robustness testing strategies will be utilized. Most of these techniques will be performed using the R programming language. Standard packages will be used in combination with custom-written code. One of the aims of the Biostatistics Core will be quantitative radiation dose reconstructions based on biological data such as cytogenetics, transcriptomics and metabolomics biomarker levels. Biomarker signature modulation by dose rate and biological system will also be studied in detail. This task will be performed by analyzing the data from experiments with different dose rate scenarios in different biological systems, including ex vivo blood vs. in vivo irradiation of the same species, and different mammalian species. Other important aims include investigation of biomarker signatures under complex radiation exposures like partial body and neutron+photon mixtures. Preliminary data suggest that analysis of cytogenetic damage (e.g. micronuclei) distributions per cell by machine learning is a promising approach for detecting partial body and neutron exposures. These methodologies will be developed and refined further for the three biomarkers types we have developed. The Biostatistics Core will also be involved in prediction of radiation-induced late lung injury (pneumonitis), and in quantification of biomarker correlations with post-irradiation blood count data. Accomplishing these aims is important for understanding the connections between the biomarkers measured by the Projects and the ultimate radiation-induced adverse outcomes like organ injury and death. The Core will actively seek to stimulate synergy between all Projects by developing methods for integrating cytogenetics, transcriptomics, and metabolomics data, using the complementary strengths of these different biomarker classes (e.g. different times-to-result, different dose response shapes such as concave vs. convex). Decision trees will be utilized to determine which combination of biomarkers and biodosimetry approaches is optimally useful for dealing with mass casualty events involving ionizing radiation.