| ECCB 2002 Poster sorted by: Author | Number Next | Previous poster (in order of the view you have selected) |
Title: Detection of a Sequence of Manipulated Experiments for Bayesian Learning | P122 |
Pournara, Iosifina; Wernisch, Lorenz i.pournara@cryst.bbk.ac.uk Dept of Crystallography, Birkbeck College, University of London |
| Recently, there is an increasingly interest in constructing genetic networks using DNA microarray data. A genetic network gives an insight into the complex relationships between genes and the ability to predict or generate new hypotheses for experimental testing. Bayesian Learning is one of the most popular methods for the construction of such networks. Bayesian Networks are probabilistic models that give a graphical representation of the conditional dependencies between random variables. A network is constructed and the posterior probability of this network given the data is calculated. Observational data, data obtained without any interference in the system are usually used to identify genetic networks. However, it has been shown that experimental data, data obtained by manipulating one or more variables can significantly improve the prediction of the network by indicating causal relationships. Constructing such experiments can be costly and time consuming, thus the question is which sequence of experiments should take place that will give the highest information about the underlining genetic network? An algorithm has been developed to suggest a minimum sequence of manipulated experiments that lead to the underlining genetic network. A number of experiments have been performed which show the advantages of the proposed algorithm compared to a random selection of manipulated experiments. |
| [1] Heckerman D. et al. Learning Bayesian Networks: The Combination of Knowledge and Statistical Data. Machine Learning. 20:197-243, 1995. [2] Cooper G. and Yoo C. Causal Discovery from a Mixture of Experimental and Observational Data. UAI, 1999. |