Granada 13 - 14 February, 2014
A major goal of the scientific activity is to model real phenomena by studying the dependency between entities, objects or more in general variables. Sometimes the goal of the modeling activity is simply predicting future behaviors. Sometimes the goal is to understand the causes of a phenomenon (e.g. a disease). Finding causes from data is particular challenging in bioinformatics where often the number of features (e.g. number of microarray probes) is huge with respect to the number of samples. In this context, even when experimental interventions are possible, performing thousands of experiments to discover causal relationships between thousands of variables is not practical. Dimensionality reduction techniques have been largely discussed and used in bioinformatics to deal with the curse of dimensionality. However, most of the time these techniques focus on improving prediction accuracy, neglecting causal aspects. This talk will discuss a major open issue : may feature selection techniques be useful also for causal feature selection? Is prediction accuracy compatible with causal discovery? Some results based on an information theory approach will be used to illustrate the issue.