| Abstract: |
To allow for a comprehensive and easy-to-grasp exploratory analysis of microarray data, we propose to subject the data to a multidimensional unfolding analysis. This is a geometric modelling technique that jointly maps genes and conditions in a low-dimensional space such that the distances between the points representing genes and conditions are monotonically related to the expression levels: the closer a gene is located to a condition, the higher its expression in that condition. Unfolding configurations are easy-to-grasp because they rely on a distance-based representation; they are comprehensive because not only the relation between a gene and the conditions is depicted, but also between genes and between conditions: genes that are close by have similar expression profiles while conditions that are close by elicit the same response from the genes. An additional useful feature of multidimensional unfolding is that it finds the optimal monotone transformation either per gene or per condition. In this presentation, first we present an algorithm that is convergent, computationally efficient, and that avoids degeneracies by the use of configuration restrictions; second, we apply the algorithm to microarray data; and third, we discuss an extension of the unfolding method to a combination with clustering. |
