Bayesian models for cancer prognosis by integrating diverse types of data

描述（由申请人提供）：癌症预后的新策略是基于不同来源的综合信息，包括患者的传统临床和人口学信息，如年龄、分级、肿瘤大小等，以及最近出现的基因或蛋白质标记物表达等遗传信息。实施这种策略需要有效的定量模型，将临床测量和遗传测量结合起来进行预后。该应用程序的长期目标是提高癌症预防、诊断和预后的风险预测、治疗选择和亚型分类。短期目标是通过开发创新的统计模型来提高对癌症患者治疗反应的预测，该模型整合了三种不同类型的数据，包括两种信息学数据，即蛋白质通路数据和高通量蛋白质表达数据，以及第三种类型，即标准的临床和人口统计学数据。我们将通过追求以下五个具体目标来实现该应用程序的目标：1)开发贝叶斯参数模型，将已知的遗传途径与高通量蛋白质表达测量相结合。2)建立贝叶斯非参数模型，将多种遗传途径与蛋白质表达测量相结合。3)在前两个目标提出的贝叶斯模型的基础上发展贝叶斯分类过程。4)将临床和人口统计学测量整合到贝叶斯模型中，并使用包含500多名白血病患者的蛋白质表达测量和临床测量的综合数据集应用贝叶斯分类程序。5)通过进行生物学实验来验证统计结果，这将由我们的合作生物学家完成。该研究有望为肿瘤学家提供基于综合信息的定量预后工具。这项研究的影响将是显著的，因为在该应用中开发的模型可以应用于各种癌症类型，从而有可能改善不同类型癌症患者的预后。

DESCRIPTION (provided by applicant): A new strategy in cancer prognosis is to base the decision on integrated information from different sources, including the traditional clinical and demographic information of patients, such as age, grade, and tumor size, etc, and the recently emerged genetic information like expression of gene or protein markers. Implementation of such a strategy requires efficient quantitative models that integrate the clinical measurements and genetic measurements together for prognosis. The long-range goal of this application is to improve risk predication, treatment selection, and subtype classification in cancer prevention, diagnosis, and prognosis. The short-term objective is to improve prediction of treatment response for cancer patients by developing innovative statistical models that integrate three different types of data, including two subtypes of informatics data, namely protein pathway data and high-throughput protein expression data, and a third type, which is the standard clinical and demographic data. We will accomplish the objective of this application by pursuing the following five specific aims: 1) Develop Bayesian parametric models that integrate a known genetic pathway with high-throughput protein expression measurements. 2) Develop Bayesian nonparametric model that integrate multiple genetic pathways with protein expression measurements. 3) Develop Bayesian classification procedures based on the Bayesian models proposed in previous two aims. 4) Integrate clinical and demographic measurements into the Bayesian models and apply the Bayesian classification procedures using a comprehensive data set that contains protein expression measurements and clinical measurements for more than 500 patients with leukemia. 5) Validate statistical findings by performing biological experiments, which will be done by our collaborating biologists. The proposed research is expected to provide quantitative prognostic tools for oncologists based on integrated information. The impact of the proposed research will be significant because models developed in this application can be applied to various cancer types and thus potentially improve the prognosis for patients with different types of cancer. PUBLIC HEALTH RELEVANCE: Integrating the protein expression data, the protein pathway data, and the clinical data is expected to significantly improve medical decision making such as treatment selection. The improved decisions are expected to improve the overall patient care. For example, by accurately predicting that certain treatment will not be effective for a cancer patient, this patient will no longer waste time trying out the treatment and will have a better chance finding some other more effective therapies.