A general language for networks

Network models have great potential to uncover complex interactions and pathways not visible by other methods, but network modelling is still an ad hoc activity, with no general theory to relate different approaches, In Depth: watch a web presentation of this work nor principles for determining what a network tells us about the biology that created it. Progress in data-driven ‘network biology’ has been hampered by the lack of a general language for integrating different network methods from different domains of knowledge.

To overcome these obstacles, I am developing a grammar of network methods, an empirical grammar of data manipulation extended into graph theory. The grammar is influenced by the statistical graphics work of Wilkinson¹¹ See The Grammar of Graphics and Wickham²² See A Layered Grammar of Graphics and Tidy Data, and allows us to describe the structure of different network methods (like the syntax in a human grammar) and relate them directly back to the biological data used to create the network, in order to uncover biological meaning (semantics). Using simple graph elements and a declarative syntax, the grammar is able to describe existing network modelling in biology using a common set of concepts, focused particularly on network inference from genomic, proteomic and metabolomic data.

The grammar of network methods is a new research direction stemming from work started during my PhD, where I integrated diverse mathematical and statistical methods of reverse-engineering transcriptomic networks³³ The initial computational framework was written in MATLAB, and applied these methods to generate and validate experimental hypotheses in transformed cell lines⁴⁴ Applications in melanoma prognosis and endothelial cell biology. The grammar develops this work by describing these existing analysis methods using a common language, and showing clearly their similarities and differences.

Specifically, my current research implements the grammar in two programming languages common in systems and computational biology (R and Python), and uses it to answer two major open questions in computational biology: the unification of ‘knowledge-based’ and ‘data-driven’ network methods, and the effect of ensemble network methods (the ‘wisdom of crowds’ effect).

Completing this work will transform network modelling in biology, from an ad hoc activity where different methods exist in isolation from one another, into a principled, evidence-driven activity with heuristics for which methods are fruitful on which data for which outcomes.

Featured publications

Hurley, D. G., Cursons, J., Wang, Y. K., Budden, D. M., Print, C. G., & Crampin, E. J. (2014). NAIL, a software toolset for inferring, analyzing and visualizing regulatory networks. Bioinformatics, 31(2), 277–278. doi:10.1093/bioinformatics/btu612

Hurley, D., Araki, H., Tamada, Y., Dunmore, B., Sanders, D., Humphreys, S., … Print, C. G. (2012). Gene network inference and visualization tools for biologists: application to new human transcriptome datasets. Nucleic Acids Research, 40(6), 2377–2398. doi:10.1093/nar/gkr902