Cancerclass R Package for Diagnostic Test Validation

This document is a technical report detailing the Cancerclass R package, which is designed for the development and validation of diagnostic tests using high-dimensional molecular data. It outlines the challenges associated with high-dimensional data, such as model overfitting and performance overestimation, and presents a multiple random validation protocol to address these issues. The report describes the process of feature selection, predictor construction using the nearest-centroid method, and evaluation of predictor accuracy through various validation techniques. Additionally, it illustrates the functionality of Cancerclass using two sets of cancer gene expression data, focusing on leukemia and breast cancer. The report includes detailed methodologies for calculating and visualizing prediction scores, as well as ROC analysis for assessing sensitivity and specificity in predicting distant metastases. The findings emphasize the importance of validation in achieving reliable diagnostic outcomes.