基于偶合人工神经网络的近红外光谱对D5S818基因座的分型

Genotyping of D5S818 locus based on near infrared spectroscopy and two combined artificial neural networks

本文基于近红外光谱结合化学模式识别中的偶合人工神经网络(ANN-ANN)法研究了一种快速、简单和低成本检测STR基因型的方法。选择STR基因座D5S818的差异较小的10-11、11-11、11-12与11-13基因型作为研究对象,将这四个基因型样本进行标准的PCR扩增并采集PCR产物的近红外光谱,以近红外光谱为判别变量,以其中一个ANN(rich-information-extracted ANN,RIE-ANN)用于提取建立判别模型的富信息变量,另一个ANN(discriminant-model-built ANN,DMB-ANN)即用于模型建立。ANN-ANN的网络结构为:338-30-338(RIE-ANN)和30-8-4(DMB-ANN)。对于校正集的校正均方根误差为0.0148,预测集为0.0127,预测准确率达到100%。成功实现了基于近红外光谱对STR基因型的快速、简单和低成本检测。

This paper has established a method which could be used to detect short tandem repeat（STR） rapidly, simply and low- costly based on near infrared （NIR）spectroscopy and two combined artificial neural network （ANN-ANN）. Genotypes 10-11,11-11, 11-12 and 12-13 of STR locus D5S818 were selected as the objective subjects. Through optimal polymerase chain reaction and NIR spectrum measurement, NIR spectra of samples for each genotype were obtained. For the genotype discriminant model of ANN- ANN, an ANN was used to extract the variables of rich information for establishing the model, and another ANN was used to estab- lish the genotype discriminant model The optimum structure of the established ANN-ANN is 338-30-338 for RIE-ANN and 30-8-4 for DMB-ANN. The root mean square error is 0. 0148 for the calibration sample set,and 0. 0127 for the prediction sample set. The accuracy was up to 100%. Based on NIR spectra,the paper achieves the detection of STR genotypes rapid, simple and low-cost.