Integrated Cancer Modeling: A New Dimension

点击翻译按钮获取中文摘要

Project Summary Significant effort has been devoted to cancer modeling, exploring statistical models that can more accurately describe cancer outcomes/phenotypes. Taking advantage of data collected at Yale University and other institutes and publicly available data (especially TCGA), we have conducted extensive research on integrated cancer modeling using various types of omics data. In the clinical practice of most if not all cancers, imaging techniques have been routinely used. Specifically, radiologists use CT/MRI/PET and other techniques, generate radiological images, and report the “macro” features of tumors. Then with samples retrieved via biopsy, pathologists review the slides of representative sections of tissues and make definitive diagnosis – this procedure generates pathological (diagnostic) images, which contain rich information on the “micro” features of tumors. Omics and pathological imaging data have been separately analyzed and established as highly effective for cancer modeling. However, a critical and practically highly relevant question, which remains unanswered, is “for more accurate cancer modeling, is it necessary to integrate omics and pathological imaging data?”. Our ultimate goal is to more effectively model cancer outcomes/phenotypes by integrating multiple sources/types of data, so as to advance cancer research and practice. In this study, we will take advantage of data recently collected under multiple Yale studies and TCGA, significantly expand the integrated analysis paradigm developed for omics data, and innovatively integrate various types of omics and pathological imaging data for cancer modeling. Three highly interconnected aims have been designed to comprehensively and complementarily study different perspectives of data integration. Aim 1: Assess the degree of overlapping information in cancer-associated omics and imaging features/models. This analysis will reveal whether overall omics and imaging data contain independent information and its degree, which is the foundation of data integration. Aim 2: Identify individual imaging (omics) features that are independently associated with cancer beyond omics (imaging) features. This analysis will identify the most important imaging/omics features, which are likely to have the highest scientific, clinical, and statistical value. Aim 3: Construct integrated models using all omics and imaging features. This analysis can lead to “mega” models, which are superior to those constructed using omics and imaging data separately, as well as rigorous measures of improvement. Such results will have high clinical relevance. This study will deliver an innovative analysis pipeline and multiple novel methods for integrating omics and pathological imaging data. With omics and pathological imaging data now routinely collected in cancer research and practice, this study will open a new venue and have a high and long-lasting impact.