Exploiting the geometry of gene expression patterns for unsupervised learning

Typical gene expression clustering algorithms are re-stricted to a specific underlying pattern model while over-looking the possibility that other information carrying pat-terns may co-exist in the data. This may potentially lead to a large bias in the results. In this paper we discuss a new method that is able to cluster simultaneously various types of patterns. Our method is based on the observation that many of the patterns that are considered significant to in-fer gene function and regulatory mechanisms all share the geometry of linear manifolds.