Integrating gene expression profiling and clinical data

We propose a combination of machine learning techniques to integrate predictive profiling from gene expression with clinical and epidemiological data. Starting from BioDCV, a complete software setup for predictive classification and feature ranking without selection bias, we apply semisupervised profiling for detecting outliers and deriving informative subtypes of patients. During the profiling process, sampletracking curves are extracted, and then clustered according to a distance derived from dynamic time warping. Sampletracking allows also the identification of outlier cases, whose removal is shown to improve predictive accuracy and stability of derived gene profiles. Here we propose to employ clinical features to validate the semisupervising procedure. The procedure is demonstrated in the analysis of a liver cancer dataset of 213 samples described by 1993 genes and by pathological features.