This article is part of the supplement: Eighth International Conference on Bioinformatics (InCoB2009): Computational Biology
MINER: exploratory analysis of gene interaction networks by machine learning from expression data
- Equal contributors
1 School of Computer Science and Engineering, The University of New South Wales, Sydney, NSW, 2052, Australia
2 School of Biotechnology and Biomolecular Sciences, The University of New South Wales, Sydney, NSW, 2052, Australia
3 The Tumour Bank, The Oncology Research Unit, The Children's Hospital at Westmead, Westmead, NSW, 2145, Australia
4 The Oncology Department, The Children's Hospital at Westmead, Westmead, NSW, 2145, Australia
BMC Genomics 2009, 10(Suppl 3):S17 doi:10.1186/1471-2164-10-S3-S17Published: 3 December 2009
The reconstruction of gene regulatory networks from high-throughput "omics" data has become a major goal in the modelling of living systems. Numerous approaches have been proposed, most of which attempt only "one-shot" reconstruction of the whole network with no intervention from the user, or offer only simple correlation analysis to infer gene dependencies.
We have developed MINER (Microarray Interactive Network Exploration and Representation), an application that combines multivariate non-linear tree learning of individual gene regulatory dependencies, visualisation of these dependencies as both trees and networks, and representation of known biological relationships based on common Gene Ontology annotations. MINER allows biologists to explore the dependencies influencing the expression of individual genes in a gene expression data set in the form of decision, model or regression trees, using their domain knowledge to guide the exploration and formulate hypotheses. Multiple trees can then be summarised in the form of a gene network diagram. MINER is being adopted by several of our collaborators and has already led to the discovery of a new significant regulatory relationship with subsequent experimental validation.
Unlike most gene regulatory network inference methods, MINER allows the user to start from genes of interest and build the network gene-by-gene, incorporating domain expertise in the process. This approach has been used successfully with RNA microarray data but is applicable to other quantitative data produced by high-throughput technologies such as proteomics and "next generation" DNA sequencing.