Current microarray experiments allow the levels of activity of thousands of genes to be measured at once, providing a window into molecular events underlying health and disease.
The selection of genes having distinct levels of activity between conditions of interest (such as cancer and non-cancer) has therefore emerged as a key aim of data analysis. However, with typically many thousands of genes to choose from and at most a few dozen sets of measurements available, differential analyses of this kind are extremely challenging. Different statistical tests yield different results due to their underlying assumptions, but on real data it is usually impossible to tell which method is likely to be right.
Researchers at the University of Oxford have developed a new method that is able of provide a consistency measure for such tests. It is capable of assessing the effectiveness of each algorithm for particular data and it can be further utilised to learn how to produce an effective statistical method for testing the given data.
The new method has many distinct advantages and benefits in comparison with existing methods for screening. One of the main advantages is that it is able to assess statistical algorithms by selecting custom algorithms from data using a notion of consistency. The technique, which is extremely robust, helps to reduce the risk of choosing an inappropriate algorithm. This helps to minimise errors and therefore lead to significant potential reduction in the cost of producing data.