人工神经网络-紫外光谱法短串联重复序列基因分型检测  

Genotyping of Short Tandem Repeat Based on Ultraviolet Spectroscopy Combined with Artificial Neural Network

作　　者：汪雪娇[1] 牟红元[1] 鲁辉[1] 豆兴茹[1] 邱婷 谢洪平[1] 

机构地区：[1]苏州大学医学部药学院,苏州215123

出　　处：《分析化学》2011年第11期1701-1705,共5页Chinese Journal of Analytical Chemistry

摘　　要：以STR基因座D16S539中的总核心重复串数相差较小的l伊11，10-12，11-11和10-13基因型为研究对象，以紫外光谱为判别变量，建立了以人T神经网络（ANN）提取富信息变量为基础的ANN基因分型方法。在优化条件下，埘4个基因型样本进行了聚合酶链式反应扩增，以扩增样本在200310nm范围内的检测光谱进行预处理和偶合的ANNANN网络优化。结果表明，提取富信息变量和基因分型的ANN的最优网络结构分别为391—50-391和50-6-4，该结构下的判别模型的校正相对均方根误差（RMS）和预测RMS分别是0．0279和0．0418，模型表现出了良好的稳健性和100％的基因型正确预测率。成功实现了基于紫外光谱对STR基因型的快速、简单和低成本检测。Taking genotypes 10-11, 11-11, 10-12 and 10-13 of short tandem repeat （STR） locus D16S539 commonly used in forensic medicine as study objects, an ANN genotyping method was developed based on ultraviolet spectra of the measured samples and another ANN was used to extract the variables of rich information. Under the optimal conditions, each of the genotypes was amplified. The ultraviolet spectra of the samples that were produced by polymerase chain reaction, which was measured at length range of 200-310 nm, were pretreated and optimized by coupled ANN-ANN. The results showed that the best network structures of the rich information extraction ANN and the discriminant model built ANN were 391-50-391 and 50-6-4, respectively. The root mean square error for the training and the prediction samples sets was obtained to be 0. 0279 and 0. 0418. It was indicated that the models had a good ability of the robustness and big discriminating power for the predic tion samples （the accuracy was 100%）. The detection of STR genotypes by UVS was rapid, simple and low-cost.

关 键 词：短串联重复序列 紫外光谱 人工神经网络 基因分型 

分 类 号：R346[医药卫生—基础医学] O657.32[理学—分析化学]