OBJECTIVE: Mitochondrial genetics has unique features that impede analysis of the biological significance of mitochondrial mutations. Simple searches for differences in total mutational load between normal and pathological samples have been frequently unrewarding, raising the possibility that more complex patterns of mutations may be responsible for some conditions. We explore this possibility in the context of Parkinson's disease (PD). METHODS AND MATERIALS: We report the development of a modified genetic algorithm suited for detection of biologically meaningful patterns of mitochondrial mutations. The algorithm is applied to a database of mutations derived from biological samples, and verified by the use of shuffled data, and repeated leave-one-out testing. RESULTS: It is possible to derive, from a very small sample, multiple accurate classifier functions that correlate with biological features. The methodology is validated statistically through experiments with fabricated data. CONCLUSION: This algorithm might be generally applicable to conditions where interactions among multiple mitochondrial DNA mutations are important. The patterns embodied in the classifier functions obtained should be the subject of further experimental study.
OBJECTIVE: Mitochondrial genetics has unique features that impede analysis of the biological significance of mitochondrial mutations. Simple searches for differences in total mutational load between normal and pathological samples have been frequently unrewarding, raising the possibility that more complex patterns of mutations may be responsible for some conditions. We explore this possibility in the context of Parkinson's disease (PD). METHODS AND MATERIALS: We report the development of a modified genetic algorithm suited for detection of biologically meaningful patterns of mitochondrial mutations. The algorithm is applied to a database of mutations derived from biological samples, and verified by the use of shuffled data, and repeated leave-one-out testing. RESULTS: It is possible to derive, from a very small sample, multiple accurate classifier functions that correlate with biological features. The methodology is validated statistically through experiments with fabricated data. CONCLUSION: This algorithm might be generally applicable to conditions where interactions among multiple mitochondrial DNA mutations are important. The patterns embodied in the classifier functions obtained should be the subject of further experimental study.