基于miRNA相对表达量自动判别月经血和外周血

目的探索有效区分月经血和外周血的miRNA最优标记组合及最佳分类模型,并构建简便快速的自动化判别软件。方法对10种miRNA(miR-451a、miR-205-5p、miR-203a-3p、miR-214-3p、miR-144-3p、miR144-5p、miR-654-5p、miR-888-5p、miR-891a-5p、miR-124-3p)在200余份月经血和外周血样本中的相对表达量以实时荧光定量PCR检测,并以7种算法模型(核密度估计、K-最近邻、逻辑回归、线性判别分析、支持向量机、神经网络、随机森林)进行数据分析,选出鉴别效果最好的标记组合及算法模型,进而构建自动判别软件。结果月经血和外周血中差别最大的三种miRNA为miR-205-5p、miR-203a-3p和miR-214-3p,使用miR144-5p与上述miRNA中的一种或两种组合可达较好区分效果,其中基于miR-144-5p、miR-203a-3p和miR205-5p所形成的“最优特征项组合一”稳健性最强。7种算法模型中最佳分类模型为核密度估计模型,其次为逻辑回归模型。结论本研究建立的自动判别软件界面友好、使用简单,适合辅助法医检验关于月经血和/或外周血判别分析的计算,便利于法医物证工作,有较大的推广应用价值。

Devising Software Enabling Automatic Discrimination between Menstrual and Peripheral Blood Based on miRNA Relative Expressions

Objective To explore the optimal combination of miRNA markers and classification model for effectively distinguishing menstrual from peripheral blood so as to build up one piece of simple and fast automatic discriminant software. Methods 10 kinds of miRNAs (miR-451a, miR-205-5p, miR-203a-3p, miR-214-3p, miR-144-3p, miR-144-5p, miR-6545p, miR-888-5p, miR-891a-5p, miR-124-3p) were analyzed through quantitative real-time PCR into their relative expression quantities from menstrual (104 pieces) and peripheral (136 pieces) blood samples. Seven algorithmic models (kernel density estimation, K-nearest neighbor, logistic regression, linear discriminant analysis, supportive vector machine, neural network, random forest) were used for data analysis so that both the optimal miRNA marker combinations and appropriate algorithmic models were selected. Consequently, the software was therewith to construct for automatically distinguishing menstrual from peripheral blood with better identification effect. Results Three miRNAs of miR-205-5p, miR-203a-3p and miR-214-3p were of greatest difference between menstrual and peripheral blood, hence coming forth the better/optimal combinations of one or two of them assembling with miR-144-5p. Among the optimal combinations, the recommended one of miR-144-5p, miR203a- 3p and miR-205-5p demonstrated most robust. The appropriate classification model was the kernel density estimation for all the seven algorithmic ones, with the logistic regression being followed. Conclusions The automatic discriminant software constructed in this study is of friendly interface, simple use, accurate and reliable server algorithm, suiting for assisting forensic calculation on menstrual and peripheral blood identification, therefore capable of effectively facilitating the forensic analysis of evidential materials and great value of promotion and application. © 2021, Editorial Office of Forensic Science and Technology. All rights reserved.