Network science, the study of complex interconnected systems, has grown over the past few years as it has become pivotal in understanding a wide variety of fields ranging from molecular and cell biology to social and information sciences and big data. By studying the structure and connectivity patterns of a network, researchers are able to gain insight into how different variables within a network are related and the strength of individual relationships, as well as make predictions about emerging network properties.
In engineered networks, the connections between elements are directly observable. For most social, biological and information networks, the connections are only inferred based on the behavior of elements in the system across time or conditions.“The challenge is that some elements may appear correlated in their activity patterns solely because they are both connected to a third party,” says Manolis Kellis, an associate professor of computer science at MIT. “For example, if my son comes to several genomics conferences with me, it may seem like he is part of the genomics community, but the link is only indirect.” These indirect links have plagued the field of network science since its inception, and as data mining capabilities are able to capture increasingly subtle interactions in increasingly dense networks, the challenge is only greater.
In a new paper appearing in the August edition of Nature Biotechnology, Kellis, graduate student Soheil Feizi, and fellow researchers describe a new algorithm that can infer direct dependencies in a network. Continue reading at MIT NEWS. Photo by Soheil Feizi, Daniel Marbach, Manolis Kellis and Steven Lee.