Novel algorithms for inferring granger causality using biological data

• Main Author: Furqan, Mohammad Shaheryar
• Other Authors: Mohammed Yakoob Siyal
• Format: Theses and Dissertations
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• Online Access: http://hdl.handle.net/10356/72683
• Institution: Nanyang Technological University

Biological network diagrams provide a natural means to characterize the association between biological entities such as genes, proteins, or brain regions. The understanding of these biological networks provides a range of information from systematic behaviours to disease susceptibility and its treatment. Currently, researchers from diverse research backgrounds are trying to model biological networks using various analytical modalities to improve understanding and prediction of biological networks. However, with the advancement of technology, the inference of biological networks from high-throughput data has received immense consideration throughout the last decade, and is a major area of research in systems biology. One of the conventional methods used for inferring biological networks from time series data is Granger causality, which analyses the causal link between the variables and can be used to map the associations in the form of a network diagram using graph theory. However, when using high-throughput data, current implementations of Granger causality face limitations, especially while handling high dimensionality and non-linear relationships among correlated variables. In order to minimize these impediments, we have proposed three heuristic techniques, and two variants of bi-directional Granger causality. We call the heuristic techniques random Forest Granger causality (RFGC), Elastic-net copula Granger causality (ECGC) and random forest copula Granger causality. The bi-directional Granger causality methods include forward backward pairwise Granger causality (FCPGC) and bi-directional random forest Granger causality (BRFGC). All these techniques are capable of handling both high- and low-dimensional data. The effectiveness of these methods is demonstrated through extensive experimentation using simulated and real datasets. We also used these methods on two real world biological applications: in mapping a gene network, commonly involved in cancer using a HeLa cancer cell dataset; and in an effective brain network, normally involved in human deductive reasoning.