Gorin Igor, Balanovsky Oleg, Kozlov Oleg, Koshel Sergey, Kostryukova Elena, Zhabagin Maxat, Agdzhoyan Anastasiya, Pylev Vladimir, Balanovska Elena
Vavilov Institute of General Genetics, Russian Academy of Sciences, Moscow, Russia.
Moscow Institute of Physics and Technology, Dolgoprudny, Russia.
Front Genet. 2022 May 16;13:902309. doi: 10.3389/fgene.2022.902309. eCollection 2022.
Currently available genetic tools effectively distinguish between different continental origins. However, North Eurasia, which constitutes one-third of the world's largest continent, remains severely underrepresented. The dataset used in this study represents 266 populations from 12 North Eurasian countries, including most of the ethnic diversity across Russia's vast territory. A total of 1,883 samples were genotyped using the Illumina Infinium Omni5Exome-4 v1.3 BeadChip. Three principal components were computed for the entire dataset using three iterations for outlier removal. It allowed the merging of 266 populations into larger groups while maintaining intragroup homogeneity, so 29 ethnic geographic groups were formed that were genetically distinguishable enough to trace individual ancestry. Several feature selection methods, including the random forest algorithm, were tested to estimate the number of genetic markers needed to differentiate between the groups; 5,229 ancestry-informative SNPs were selected. We tested various classifiers supporting multiple classes and output values for each class that could be interpreted as probabilities. The logistic regression was chosen as the best mathematical model for predicting ancestral populations. The machine learning algorithm for inferring an ancestral ethnic geographic group was implemented in the original software "Homeland" fitted with the interface module, the prediction module, and the cartographic module. Examples of geographic maps showing the likelihood of geographic ancestry for individuals from different regions of North Eurasia are provided. Validating methods show that the highest number of ethnic geographic group predictions with almost absolute accuracy and sensitivity was observed for South and Central Siberia, Far East, and Kamchatka. The total accuracy of prediction of one of 29 ethnic geographic groups reached 71%. The proposed method can be employed to predict ancestries from the populations of Russia and its neighbor states. It can be used for the needs of forensic science and genetic genealogy.
目前可用的基因工具能够有效区分不同的大陆起源。然而，占世界最大大陆三分之一的北欧亚大陆，在相关研究中仍然严重缺乏代表性。本研究使用的数据集代表了来自12个北欧亚国家的266个群体，涵盖了俄罗斯广袤领土上的大部分民族多样性。总共1883个样本使用Illumina Infinium Omni5Exome - 4 v1.3 BeadChip进行基因分型。对整个数据集计算了三个主成分，使用三次迭代去除异常值。这使得266个群体能够合并为更大的群体，同时保持组内同质性，从而形成了29个在基因上具有足够可区分性以追溯个体祖先的民族地理群体。测试了几种特征选择方法，包括随机森林算法，以估计区分这些群体所需的遗传标记数量；选择了5229个具有祖先信息的单核苷酸多态性（SNP）。我们测试了各种支持多类别的分类器以及每个类别可解释为概率的输出值。逻辑回归被选为预测祖先群体的最佳数学模型。用于推断祖先民族地理群体的机器学习算法在原始软件“Homeland”中实现，该软件配备了接口模块、预测模块和制图模块。提供了显示北欧亚不同地区个体地理祖先可能性的地理地图示例。验证方法表明，在南西伯利亚和中西伯利亚、远东和堪察加半岛，观察到的民族地理群体预测数量最多，且几乎具有绝对准确性和敏感性。对29个民族地理群体之一的预测总准确率达到71%。所提出的方法可用于预测俄罗斯及其邻国人口的祖先。它可用于法医学和遗传谱系学的需求。
