Title: Function-on-function mixture model clustering
Authors: Shahin Tavakoli - University of Geneva (Switzerland)
Daphne Ezer - The University of York - The University of Warwick - Alan Turing Institute London (United Kingdom)
Susana Conde Llinares - The University of Warwick, Alan Turing Institute, London (United Kingdom) [presenting]
Abstract: Gene expression data is often collected over time in a variety of experiments under different experimental conditions. Genes may have very different temporal gene expression profiles, but adjusting their expression patterns in the same way through experimental conditions. We aim to find clusters that capture functional regression relationships between a temporal response and temporal explanatory variables, possibly more than one. We develop a K-means type iterative-consensus clustering algorithm in which each cluster is defined by a function-on-function regression model fitted using boosting. Our models allow for many situations, including even autoregressive random error terms inter alia. We validate them with extensive simulations and then apply them to identify groups of genes whose diurnal expression pattern is similarly perturbed by the season. Our clusters are enriched for genes with similar biological functions, including one characterized with both photosynthesis-related functions and polysomal ribosomes.