Multitask learning for host–pathogen protein interactions

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

The regularizer in Equation (3) uses the pathway information matrix to enforce pathway-level similarity. The matrix can be used to represent any other common structure. For example, consider the hypothesis 'all pathogens target hub proteins in the host', which implies that bacterial proteins are often found to interact with host proteins that have a high node degree in the PPI network of the host. We tried two variants to incorporate this hypothesis-(i) we identify 'hubs' in the human PPI graph and use the binary vectors p i as an indicator of the 'hub' protein targeted by the bacterial protein, (ii) instead of a discrete 'hub'/ 'not hub' indicator we use p i to represent the node degree [each component of p i represents one node-degree bin say (10-20)]. We found that using (i) gives us an improvement of upto 2.5 F points over the baseline methods.