Multitask learning for host–pathogen protein interactions

Kshirsagar, Meghana; Carbonell, Jaime; Klein-Seetharaman, Judith

Multitask learning is a family of machine learning methods that addresses the issue of building models using data from multiple problem domains (i.e. 'tasks') by exploiting the similarity between them. The goal is to achieve performance benefits for all the tasks involved. This paradigm of building joint models has been applied successfully in many areas including text mining, computer vision, etc. Because bioinformatics datasets often represent an organism, a natural notion of a 'task' is an 'organism'-for example, the work by Widmer et al. (2010) uses a multitask learning approach for splice-site prediction across many organisms. They use phylogenetic trees to incorporate similarity between organisms (i.e. tasks). For a survey of multitask learning in computational biology, see Xu and Yang (2011) .