| Detecting regulation relations between genes in a specie is one of the most important biological problems. Many attempts have been made to detect regulation relations in microarray expression data and other biological data sources using data mining techniques. These attempts were only partially successful, and the number of undiscovered relations remains large. As of today there is no known method for detecting regulation relations in microarray data.; In this dissertation three different data mining contributions for solving this problem are presented. The first is a new approach for solving the well-known Candidate set generation problem, which is a major performance bottleneck for some approaches. Second, an algorithm to efficiently pre-filter more than %95 of possible triplet regulation relations is described. Finally, a new Hidden Markov Model machine learning approach for scoring temporal relationships between gene expressions is introduced. Each of these methods has applications to a significant class of non-biological data mining problems. |
