High-confidence assessment of functional impact of human mitochondrial non-synonymous genome variations by APOGEE

24,189 are all the possible non-synonymous amino acid changes potentially affecting the human mitochondrial DNA. Only a tiny subset was functionally evaluated with certainty so far, while the pathogenicity of the vast majority was only assessed in-silico by software predictors. Since these tools proved to be rather incongruent, we have designed and implemented APOGEE, a machine-learning algorithm that outperforms all existing prediction methods in estimating the harmfulness of mitochondrial non-synonymous genome variations. We provide a detailed description of the underlying algorithm, of the selected and manually curated training and test sets of variants, as well as of its classification ability.所有可能影响人类线粒体DNA（mitochondrial DNA）的非同义氨基酸变异共计24189种。截至目前，仅有极小一部分变异完成了确定性的功能验证，绝大多数变异的致病性仅通过软件预测工具进行了计算机模拟（in silico）评估。鉴于现有工具的预测结果一致性较差，我们设计并实现了APOGEE——一款机器学习（machine learning）算法，其在评估线粒体非同义基因组变异危害性的任务中，性能优于所有现有预测方法。本研究详细介绍了该核心算法的细节、经筛选与人工整理的变异训练集与测试集，以及该模型的分类性能。