Mertens Bart J A, De Noo M E, Tollenaar R A E M, Deelder A M
Department of Medical Statistics, Leiden University Medical Centre, Leiden, The Netherlands.
J Comput Biol. 2006 Nov;13(9):1591-605. doi: 10.1089/cmb.2006.13.1591.
This paper presents an approach to the evaluation and validation of the diagnostic potential of mass spectrometry data in an application on the construction of an "early warning" diagnostic procedure. Our approach is based on a full implementation and application of double cross-validatory calibration and evaluation. It is a key feature of this methodology that we can jointly optimize the classifiers for prediction while simultaneously calculating validated error rates. The methodology leaves the size of the training data nearly intact. We present application to data from a designed experiment in a colon-cancer study. Subsequent to presentation of results from the double cross-validatory analysis, we explore a post-hoc analysis of the calibrated classifiers to identify the markers that drive the classification.
本文提出了一种方法，用于评估和验证质谱数据在构建“早期预警”诊断程序中的诊断潜力。我们的方法基于双交叉验证校准和评估的全面实施与应用。该方法的一个关键特性是，我们可以在计算验证错误率的同时，联合优化用于预测的分类器。此方法几乎不改变训练数据的规模。我们展示了该方法在一项结肠癌研究的设计实验数据中的应用。在展示双交叉验证分析的结果之后，我们对校准后的分类器进行事后分析，以识别驱动分类的标志物。
