Efficient analysis of microarray data often balances two needs: the requirement for close interaction with experimentalists and the power of advanced models.

While existing tools like limmaGUI make it easy to rerun an existing model on new data, there is no support for the definition of the model. Many efficient analyses follow a balanced design and are based on an ANOVA. Defining a mixed effects linear model and the relevant contrasts, however, is not trivial for biologists. In Sykcek et al. (2005) we have introduced a new approach to facilitate the interaction between experimentalists and data analysis experts for that task. Processing is controlled by a simple text file that defines and documents all steps of the analysis. The Friendly Statistics Package for Microarray Analysis provides the tool as well as extensive instructions. In addition, we offer hands-on lecture courses.

In Sykcek et al. (2007) we have introduced and evaluated the power of a hierarchical Bayesian model for the integrated analysis of several microarray data sets in order to identify shared gene function. The approach has been been tested in simulations as well as in the combined analysis of in vitro and in vivo experiments using well understood experimental conditions as a reference. As expected, genes associated with apoptosis were identified as implicated both in mouse mammary gland development and cell line growth factor withdrawal. Our approach identified relevant genes with higher sensitivity and specificity than traditional threshold based alternatives.

References

1. Sykacek P, Furlong RA, Micklem G (2005) A friendly statistics package for microarray analysis. Bioinformatics 21, 4069–70. (read more)
2. Sykacek P, Clarkson R, Print C, Furlong R, Micklem G (2007) Bayesian modelling of shared gene function. Bioinformatics 23, 1936–44. (read more | Supplement)

---

Return to group home or publications