| 引用本文: | 刘子铭,郭佩宏,孙永恒,祖建,胡曦,马欣越,吴晓明,王彤.染色体易位重组位点的自动识别方法研究[J].生物信息学,2021,19(3):159-169. |
| LIU Ziming,GUO Peihong,SUN Yongheng,ZU Jian,HU Xi,MA Xinyue,WU Xiaoming,WANG Tong.Automatic identification of recombination sites for translocated chromosome[J].Chinese Journal of Bioinformatics,2021,19(3):159-169. |
|
| |
|
|
| 本文已被：浏览 1757次 下载 21次 |
码上扫一扫！ |
|
|
| 染色体易位重组位点的自动识别方法研究 |
|
刘子铭1,郭佩宏1,孙永恒1,祖建1,胡曦2,马欣越3,吴晓明2,王彤4
|
|
(1.西安交通大学 数学与统计学院, 数学与生命科学交叉中心, 西安 710049; 2.西安交通大学 生命科学与技术学院,生物医学信息工程教育部重点实验室,西安 710049; 3.西安交通大学 医学部, 西安 710061;4.河北燕达陆道培医院, 河北 廊坊 065201)
|
|
| 摘要: |
| 染色体易位重组位点的识别对很多染色体遗传性疾病的诊断有着重要的意义。本文基于实际诊断中采集到的24类染色体数据和9号正常与异常染色体数据,构建了一套自动识别染色体易位重组位点的模型和方法。首先,对染色体图像进行预处理,得到了方向梯度直方图特征(HOG)和局部二值模式特征(LBP),构建了基于纹理特征的染色体24分类多通道网络模型,分类准确率达到了95.99%；再与ResNet18模型(分类准确率95.86%)进行模型融合,最终分类准确率达到97.08%。其次,将染色体密度谱作为正常和异常染色体的分类特征,采用投票的方法集成支持向量机、随机森林和XGBoost模型,构建了正常和异常染色体的集成分类器,正常和异常9号染色体的分类准确率达到了100%。最后,对于易位的异常染色体,我们提出了基于动态时间规划(DTW)的易位重组位点自动识别算法,在异常染色体的密度谱曲线上找到了重组位点,并映射至染色体G显带模式图,得到标准诊断结果,通过与临床专家的诊断结果进行比较说明了自动识别结果的有效性。本文设计的一套自动识别染色体易位重组位点的模型方法对临床辅助诊断有很大的帮助,有望完善成为一套软件系统应用于临床诊断,提升相关疾病的诊断效率和准确率。 |
| 关键词: 染色体分类 异常染色体 重组位点 自动识别 深度学习 |
| DOI：10.12113/202008007 |
| 分类号:Q343.2 |
| 文献标识码:A |
| 基金项目:国家自然科学基金项目(No.5,2)；陕西省自然科学基础研究计划(No.2019JM-478). * |
|
| Automatic identification of recombination sites for translocated chromosome |
|
LIU Ziming1, GUO Peihong1, SUN Yongheng1, ZU Jian1, HU Xi2, MA Xinyue3, WU Xiaoming2, WANG Tong4
|
|
(1. Interdisciplinary Research Center for Mathematics and Life Sciences, School of Mathematics and Statistics, Xian Jiaotong University, Xian 710049, China;2.The Key Laboratory of Biomedical Information Engineering of Ministry of Education, School of Life Sciences and Technology, Xian Jiaotong University, Xian 710049, China; 3.Health Science Center, Xian Jiaotong University, Xian 710061, China;4.Hebei Yanda Lu Daopei Hospital, Langfang065201, Hebei China)
|
| Abstract: |
| Precise identification of recombination sites of translocated chromosome is of great significance for the diagnosis of many chromosomal inherited diseases. Based on the 24 types of chromosome data and the normal and abnormal chromosome 9 data collected in actual diagnoses, a deep learning method was constructed for the automatic identification of chromosomal recombination sites. First, through image preprocessing, the histogram of oriented gradients (HOG) feature and local binary patterns (LBP) feature of chromosome images were obtained. Based on these texture features, a multi-channel classification model was constructed, and the classification accuracy reached 95.99%. By fusing the proposed model with the ResNet 18 model(classification accuracy of 95.86%),the final classification accuracy reached 97.08%. Then, the chromosome density profile was used as the classification feature of normal and abnormal chromosomes. By using voting method to integrate the results of support vector machine, random forest, and XGBoost models, an integrated normal and abnormal chromosome classifier was constructed with a classification accuracy of 100%. After obtaining the abnormal chromosome, based on the method of dynamic time warping (DTW), a DTW automatic identification algorithm of recombination site was developed, and the recombination site on the density profile of abnormal chromosome was found. In addition, the recombination site was further mapped to the G-banded karyotypes. By comparing with the results of clinical experts, the rationality of the model estimated results was verified. The automatic identification method of chromosome recombination sites designed in this paper is of great significance for clinically assisted diagnosis. It is expected to be designed as a software for clinical diagnosis so as to improve the diagnosis efficiency of related diseases. |
| Key words: Chromosome classification Abnormal chromosome Recombination site Automatic identification Deep learning |
|
|
|
|
