Basically, this paper presents a mitochondrial phenotypic database (http://www.mitophenome.org/) and carried some further analysis based on it.
Recent study showed that mitochondrial related dysfunctions are more prevalent in hereditary diseases than previously anticipated. But, in literature or database, there is no standard formats of such phenotypic data of mitochondrial disease
And there is no standard catalog of the associated clinical phenotypes.
Firstly, the author manually identified individual clinical features, signs and symptoms of 174 mitochondrial genes, which are associated with 191 disease in the OMIM. Then, a manual annotation process was carried, which is consisted of the collection, definition and classification of phenotypic features.
The figure1 is a categorical breakdown of the collected 502 features in their fourteen clinical categories. From this figure, we can see, the neurologic and metabolic CC contained the largest fraction of features (18.5% and 14.3%, respectively). Interestingly, this distribution of phenotypes within CC is largely consistent with the tissue distribution of energy expenditure in the resting state, or basal metabolic rate (BMR).
The author proposed Quantitative Phenotypic Associations (QPA) as a quantitative measure of phenotype similarity of disease genes causing one or more identical phenotypic feature.
The analysis show that the association between QPA and Likelihood Ratios is positive.
Then same methods was expanded to identify functional interactions among 162 disease genes and 4577 candidates genes. This involves 1.9 million interactions between DG-DG, DG-CG, CG-DG and CG-CG. Then identified 495 mitochondrial CG through data integration.
Another analysis was done in this paper is to predict functional candidate genes. Firstly, the author used a supervised discriminant analysis of all 695 mitochondrial genes using the five attributes of gene functional interactions. Besides the 495 mitochondrial CG, 254 genes were predicted as DG with a true positive rate of 80.2% based on the confirmed known DG. In addition, 26 of the 38 DG with likely mitochondrial localization, which were input-labeled as CG to serve as controls, were correctly classified as DG. As an alternative tool, the author ran a supervised Bayesian network approach. They first defined a training set of 100 typical out of the 162 mitochondrial DG based on their median of total gene interactions. Accordingly, 100 typical CG were selected from the 495 mitochondrial CG. The network analysis correctly identified 56.8% of the DG, 16 out of the 38 likely mitochondrial DG, and predicted 201 DG out of the 495 CG. Overlapping the two approaches showed its potential of predictions as well. It predicted 168 novel mitochondrial DG with an estimated true positive rate of 85.8% (139 out of 162 DG) based on the correctly classified DG (Table S8).
In this study, a knowlegebase was created for storing detailed phenotypic information of known mitochondrial genes.
Then some methods were developed to analyze the clinical phenotype information, (1) to determine associations of genes and diseases, (2) compare different disease genes based on their associated phenotypes. Ultimately, this approach was used to predict disease gene similarities, which showed positive correlations to their functional interactions. By which, to predict disease candidate genes.
