Entropy network fusion

Abstract

Thanks to the continual development of technology, a massive amount of data is now being produced on a daily basis. Because of this, new methods for analysis are needed, particularly ones that can analyse multiple datasets on the same set of objects. This is especially relevant in systems biology, where different datasets probe different aspects of the same underlying system. A popular and widely used method to analyse datasets is to transform the data into networks. Networks help visualise and quantify connections between samples, revealing structure and information that may not be visible at first, hence the popularity of their use, particularly in the analysis of biological systems. Entropy Network Fusion (ENF) is a new methodology for fusing, or combining, together multiple networks on the same set of objects (nodes) into one single output network. It
works by finding a solution (network) whose clustering structure is as close as possible to the clustering structure of all the given input networks, using information-theoretic entropy as a guiding principle. ENF is designed with a level of generality, such that it is not restricted to any specific type of data, giving it a wide range of applications. We tested our methodology on five cancer sets and compared the performance to Similarity Network Fusion, a state-of-the-art network fusion algorithm. Whilst SNF may be a faster method, the output from ENF is significantly better in terms of performance. We then further developed an approximate version of our algorithm, approximate Entropy Network Fusion (aENF), which is significantly faster computationally for larger networks, further increasing its range of application.

More information

Identifiers

ORCID for Ruben Sanchez Garcia: orcid.org/0000-0001-6479-3028

Catalogue record

Export record