Link discovery in graphs derived from biological databases

Public biological databases contain vast amounts of rich data that can also be used to create and evaluate new biological hypothesis. We propose a method for link discovery in biological databases, i.e., for prediction and evaluation of implicit or previously unknown connections between biological entities and concepts. In our framework, information extracted from available databases is represented as a graph, where vertices correspond to entities and concepts, and edges represent known, annotated relationships between vertices. A link, an (implicit and possibly unknown) relation between two entities is manifested as a path or a subgraph connecting the corresponding vertices. We propose measures for link goodness that are based on three factors: edge reliability, relevance, and rarity. We handle these factors with a proper probabilistic interpretation. We give practical methods for finding and evaluating links in large graphs and report experimental results with Alzheimer genes and protein interactions.