Bioinformatics analysis of mitochondrial disease

Abstract

Several bioinformatic methods have been developed to aid the identification of novel nuclear-mitochondrial genes involved in disease. Previous research has aimed to increase the sensitivity and specificity of these predictions through a combination of available techniques. This investigation shows the optimum sensitivity and specificity can be achieved by carefully selecting seven specific classifiers in combination. The results also show that increasing the number of classifiers even further can paradoxically decrease the sensitivity and specificity of a prediction. Additionally, text mining applications are playing a huge role in disease candidate gene identification providing resources for interpreting the vast quantities of biomedical literature currently available. A workflow resource was developed identifying a number of genes potentially associated with Lebers Hereditary Optic Neuropathy (LHON). This included specific orthologues in mouse displaying a potential association to LHON not annotated as such in humans. Mitochondrial DNA (mtDNA) fragments have been transferred to the human nuclear genome over evolutionary time. These insertions were compared to an existing database of 263 mtDNA deletions to highlight any associated mechanisms governing DNA loss from mitochondria. Flanking regions were also screened within the nuclear genome that surrounded these insertions for transposable elements, GC content and mitochondrial genes. No obvious association was found relating NUMTs to mtDNA deletions. NUMTs do not appear to be distributed throughout the genome via transposition and integrate predominantly in areas of low %GC with low gene content. These areas also lacked evidence of an elevated number of surrounding nuclear-mitochondrial genes but a further genome-wide study is required.