Mining gene microarray data with adaptive feature scaling

We propose a new technique, adaptive feature scaling (AFS), to improve the performance of clustering algorithm applied to gene microarray data. In AFS, every feature is assigned multiple weights, each for an individual cluster, and the weights are adaptively updated during the clustering process so that certain features (signals) are strengthened while others (noises) are diminished. Clustering with AFS results in low-noise clusters focuses on a small set of signal features. Moreover, the contribution of each feature to each cluster can be revealed by using different feature weights. We apply AFS in conjunction with the message passing clustering (MFC) algorithm to colon cancer data set to show the potential use of AFS in genetics research and medical diagnosis