Abstract: Application of genetic algorithms to the analysis of microarray gene expressison data

New techniques of measuring the activity of thousands of genes simultaneously are revolutionizing research in biology and medicine. The commonly used analytical methods for grouping genes to distinguish between different kinds of samples are not, however, informative for finding direct relationships or simple rules for prediction.

After performing a survey of the methods used in research so far, a different approach was tried in this thesis. This approach employs a method for finding explicit relationships between genes, that can be used directly for predicting phenotypes or as a classification tool. The rules may also be used as a pool of possible relationships to test further.

The method, based on genetic algorithms, is non-deterministic and states the result as rules consisting of a set of conditions based on the expression averages of some genes. The search is conducted with a population of rules that are subjected to an artificial evolutionary process until the conditions can be met and the samples are correctly classified. The expression values used as input to the search algorithm are trinerized, as the significance of the exact level of the absolute expression values and ratios is unclear and much debated.


Magnus Alm Rosenblad, June 2001
Last modified: Wed Sep 12 12:22:02 MET DST 2001