Because the complexity of gene networks is potentially huge, the basic idea would be to implement some kind of evolutionary learning system that can evolve control of a probabilistic Boolean network, which could then be experimented with in relation to desired outcomes (for some GSK/externally derived networks/problems). This paper, http://www.aporc.org/LNOR/9/OSB2008F04.pdf , is the only one I've found that's remotely related evo-methods to this area (although things like HMM/Dynamic Programming have been used for small networks- see link), although I sense their primary context may be a little different (they seem to be modelling the system whereas we would wish to apply intervention to indirectly promote modulation of a disease-target).
We have various scenarios where it would be useful to have supervised/unsupervised learning capability for predicting functional linkages between attributes in different kinds of data and the ability to use this in predicting outcomes (one very new area would be in applying a combination of state-of-the-art epigenetic short-sequencing data with related micro-array data toward understanding the biology of DNA expression in a specific experiment). For this I had in mind (composite or) multiple-kernel learning (i.e. http://www.cs.ucl.ac.uk/staff/rmartin/smls09/posters/campbell_ying.pdf) , or maybe something like this?http://www.sciencedirect.com/science?_ob=MImg&_imagekey=B6V15-4TDVM74-1-19&_cdi =5665&_user=2529830&_orig=search&_coverDate=01%2F01%2F2009&_sk=999699998&view=c &wchp=dGLbVlb-zSkzS&md5=d43be9378f0a9880dc47e4bcd51844d0&ie=/sdarticle.pdf )
Page maintained by Xin Yao.