UC San Diego

Independent Component Analysis (ICA) is an unsupervised machine learning algorithm which models a complex multivariate dataset as a linear combination of statistically independent hidden factors. Applied to high-quality gene expression data from E. coli, it effectively reveals these hidden factors of the transcriptional regulatory network as sets of co-regulated genes and their corresponding activities across diverse growth conditions. The two main variables affecting the output of ICA are the data itself and the user-defined number of components to compute. In this study, a new method for effectively setting dimensionality is proposed which aims to maximize the number of biologically relevant components revealed while minimizing the potential for over-decomposition.