NCC-AUC: Nearest Centriod Classifier for AUC valuation

• Version 0.0.1
• Last updated: Feb. 10, 2015

References#

• Meng Zou, Zhaoqi Liu, Xiang-Sun Zhang, Yong Wang, NCC-AUC: an AUC-based method to identify multi-biomarker panel for cancer prognosis from genomic and clinical data. In submission.

Motivation#

In prognosis and survival studies, an important goal is to identify multi-biomarker panels with predictive power using molecular characteristics or clinical observations. Such analysis is often challenged by censored, small-sample-size, and high-dimensional clinical data or genomic profiles. Therefore, sophisticated models and algorithms are in pressing need to solve these issues.

Method#

Here, we propose a novel Area Under Curve (AUC)-based multi-biomarker panel identification method called NCC-AUC (Nearest Centroid Classifier for AUC evaluation). Our method is motived by the connection between AUC score in classification accuracy evaluation and Harrell’s concordance index in survival analysis. This connection allows us to convert the survival time regression problem to a binary classification problem. Then an optimization model is formulated to directly maximize AUC and meanwhile select a small group of features to construct a predictor. Specifically, NCC-AUC utilizes the nearest centroid classifier framework and optimizes the classification accuracy AUC. Simultaneously, NCC-AUC minimizes the number of features to regularize the classifier. NCC-AUC obtains a good performance on not only gene expression data of breast cancer but also clinical data of stage IB NSCLC (Non-Small-Cell Lung Cancer).

Software#

This version of the program is in very preliminary stage and provided just for testing purpose. The program is still under development.

• NCC-AUC.rar