Identification of complex functional relations among genes through association rule learning
Riccardo Percudani, Nicola Doniselli, Alessandro Dal Palù, Pietro Cravedi · 2015
The analysis of correlated and anti-correlated phylogenetic profiles is a widely used method to predict functional associations among genes. As previously pointed out, these methods corresponds to pairwise logic implications between genes of the type A → B or A → not B, respectively, i.e. the presence of gene A in a genome implies the presence of B, or the presence of gene A implies the absence of B. However, for the presence of analogous proteins, alternative pathways, and pathway branching points, the evolutionary and functional associations among genes can be far more complex than pairwise relations (1). No general algorithms have been implemented for the discovery of complex logic implications involving three or more genes. Here we describe the adaptation of the widely used Apriori algorithm of association rule learning (2,3) to the identification of complex patterns of gene presence/absence across genomes. As a proof of principle of the application of this data mining technique, we report on the identification of a missing gene in purine catabolism through the identification of an association rule relating six different genes involved in the pathway. 1. Bowers, et. al. Science (80). 306, 2246–2249 (2004). 2. Agrawal, R. & Srikant, R. in Proc. 20th Int. Conf. Very Large Data Bases 487–499 (1994). 3. Borgelt, C. Wiley Interdiscip. Rev. Data Min. Knowl. Discov. 2, 437–456 (2012).