Improving network topology-based protein interactome mapping via collaborative filtering

High-throughput screening (HTS) techniques enable massive identification of protein-protein interactions (PPIs). Nonetheless, it is still intractable to observe the full mapping of PPIs. With acquired PPI data, scalable and inexpensive computation-based approaches to protein interactome mapping (PIM), which aims at increasing the data confidence and predicting new PPIs, are desired in such context. Network topology-based approaches prove to be highly efficient in addressing this issue; yet their performance deteriorates significantly on sparse HTS-PPI networks. This work aims at implementing a highly efficient network topology-based approach to PIM via collaborative filtering (CF), which is a successful approach to addressing sparse matrices for personalized-recommendation. The motivation is that the problems of PIM and personalized-recommendation have similar solution spaces, where the key is to model the relationship among involved entities based on incomplete information. Therefore, it is expected to improve the performance of a topology-based approach on sparse HTS-PPI networks via integrating the idea of CF into it. We firstly model the HTS-PPI data into an incomplete matrix, where each entry describes the interactome weight between corresponding protein pair. Based on it, we transform the functional similarity weight in topology-based approaches into the inter-neighborhood similarity (I-Sim) to model the protein-protein relationship. Finally, we apply saturation-based strategies to the I-Sim model to achieve the CF-enhanced topology-based (CFT) approach to PIM.