基于近红外光谱的人工神经网络研究STR基因座分型方法  

Genotyping of STR Locus Based on Near-infrared Spectroscopy and Artificial Neural Network

作　　者：袁高林[1] 任丽[1] 高玉振[2] 汪维鹏[1] 鄂翔[1] 谢洪平[1] 

机构地区：[1]苏州大学医学部药学院,江苏苏州215123 [2]苏州大学医学部法医学系,江苏苏州215123

出　　处：《分析测试学报》2009年第11期1245-1249,共5页Journal of Instrumental Analysis

基　　金：国家大学生创新性实验计划资助项目(57315425)

摘　　要：以D16S539基因座的3种(9-9、9-11、11-11)基因型为例,设计引物扩增包含该多态性位点的1段DNA片段,获得了3种基因型建模样本各50个。基于近红外光谱(NIRS)结合误差反向传播人工神经网络(BPANN)建立了测定短串联重复序列(STR)基因型的判别模型,所建立的判别模型的校正均方根残差和预测集均方根误差分别为0.082 5、0.072 5,预测准确率均为100%。该方法不需任何前处理,只需一步PCR扩增和NIRS检测即可实现STR基因型判别,具有简单、快速、低成本等优点。Taking three genotypes 9 - 9, 9 - 11 and 11 - 11 of D16S539 locus as example, DNA fragments containing polymorphism sites were amplified by PCR method based on a pair of primers to obtain the three-genotype samples. For each genotype, 50 of the samples were obtained, in which 50 samples were divided randomly into two sets. One set containing 34 of the samples was used as the training one and another one containing 16 of the samples as the predicting one. Based on near-infrared spectra （NIRS） of the measured samples and back propagation artificial neural network （BPANN）, the genotype discriminant model of short tandem repeat （STR） has been established. The root mean square error for the training and the prediction sample sets were obtained to be 0. 082 5 and 0. 072 5, respectively. The accuracy of prediction samples was up to 100%. The method was simple, rapid and low-cost, and could be directly applied in the analysis of three genotypes of STR without any preprocessing.

关 键 词：近红外光谱 人工神经网络 短串联重复序列 基因型 

分 类 号：O657.33[理学—分析化学] TP183[理学—化学]