基于DNA变异的中国汉族人群脱发表型推断及预测模型评估

目的 男性型脱发（male pattern baldness，MPB），又称为雄激素性脱发（AGA），是一种常见的男性脱发类型，大约80%的表型差异可以用遗传因素解释。目前的MPB遗传推断研究主要基于欧洲人群，东亚人群相关研究较少。本研究在中国人群中对欧洲人群MPB关联位点进行验证分析，并建立遗传推断模型。方法 本研究调查了486个与欧洲人群MPB相关单核苷酸多态性（SNP）位点在312名中国汉族男性中的关联性，分别使用逐步回归和Lasso回归方法对关联出的位点进行筛选。使用逻辑回归算法构建预测模型，通过十折交叉验证的方法评估。之后进一步比较了逻辑回归、k近邻分类器、随机森林、支持向量机4种常用分类器模型对MPB的预测准确性。结果 有174个SNP位点与中国汉族男性的MPB显著相关（P＜0.05）。通过不同的筛选方法，分别得到了22个SNP和25个SNP的位点集合。基于上述位点集合建立了22-SNP和 25-SNP两种逻辑回归预测模型。以AUC（ROC曲线下方的面积大小，area under curve）来衡量，两种模型对MPB预测的准确性分别为0.85和0.84；经十折交叉验证后预测准确性分别下降至0.81和0.77。当加入年龄作为预测因子后，两种模型的AUC均达到最大值0.89。从运行结果来看，逻辑回归预测模型较本研究中的其他分类器模型具有明显优势。结论 总体而言，虽然预测模型的准确性尚未达到临床期望水平，但SNP在MPB的遗传预测方面仍具备很大的潜力，可以为MPB的早期诊断、临床干预和法庭科学应用提供参考。

Phenotypic Prediction of Male-pattern Baldness in Chinese Han Population Based on DNA Variants

Objective Male pattern baldness (MPB), or androgenetic alopecia (AGA), is a common type of hair loss in men, with an estimation that approximately 80% of the phenotypic variance can be explained by genetic factors. Most prediction models were developed in European and few MPB associated (single nucleotide polymorphisms,SNPs) have been validated in East Asian population. In this study, MPB associated SNPs in European were verified in Chinese population, and MPB risk prediction models were built based on those SNP data.Methods We examined 486 genetic variants previously reported associated with MPB, and assessed their impacts on hair loss in 312 Chinese individuals. Different sets of SNPs were selected by stepwise regression and Lasso regression. Logistic regression algorithm was used to construct the prediction models and the evaluations were conducted by the method of 10-fold cross validation. We further compared the prediction accuracy among logistic regression, k-nearest neighbor classifier, random forest and support vector machine.Results 174 SNPs demonstrated significant associations with MPB (P<0.05). Among those SNP markers, 22 SNPs and 25 SNPs were selected by different screening methods. Two logistic regression model considering the genotypes of 22 and 25 SNPs demonstrate that the risk of MPB were predictable at AUC (area under curve) level of 0.85 and 0.84. Prediction accuracy was slightly reduced after performing 10-fold cross validation, 0.81 and 0.77 respectively. Moreover, the AUC of both models reaches maximum (0.89) when age was added as a predictive factor. From the running results, the logistic regression prediction model had obvious advantages.Conclusion Overall, although the accuracy obtained here has not reached a clinically desired level, our model still has great potential for genetic prediction of MPB, which may assist decision making on early MPB intervention actions and in forensic investigations.