Target identification in biological systems using network connectivity information from literature mining databases

Abstract

We address the automated drug target identification problem for pharmaceutical research. It is often the case in pharmaceutical industry to bring a new promising target to clinical trials only to find that it has serious safety concerns or lack of efficacy. A gene downstream or upstream in the pathway can be a remedy, however, finding such an alternative target using existing in-silico or bench tools can be extremely labor-intensive. Recently, increasing amounts of information and observations have been compiled from different areas of biological research and deposited on databases. In this work we propose a novel computational method to quantify indirect relationships between the objects of biological research of interest by using existing relationships from text mining databases to automate the search for novel biological targets. We applied our method to analyze 9575 proteins in Ariadne database and create a rank-ordered list of proteins that are most similar to the original query. We also compared our method with the Jaccard similarity index for link prediction performance. Our method outperformed the Jaccard method in predicting the existing links for 9575 proteins in the database.