通过机器学习鉴定亨廷顿病的致病基因。

Identification of contributing genes of Huntington's disease by machine learning.

机构信息

Graduate Institute of Integrated Medicine, College of Chinese Medicine, China Medical University, Taichung, 40402, Taiwan.

Department of Medical Research, China Medical University Hospital, Taichung, 40447, Taiwan.

出版信息

BMC Med Genomics. 2020 Nov 23;13(1):176. doi: 10.1186/s12920-020-00822-w.

BACKGROUND

Huntington's disease (HD) is an inherited disorder caused by the polyglutamine (poly-Q) mutations of the HTT gene results in neurodegeneration characterized by chorea, loss of coordination, cognitive decline. However, HD pathogenesis is still elusive. Despite the availability of a wide range of biological data, a comprehensive understanding of HD's mechanism from machine learning is so far unrealized, majorly due to the lack of needed data density.

METHODS

To harness the knowledge of the HD pathogenesis from the expression profiles of postmortem prefrontal cortex samples of 157 HD and 157 controls, we used gene profiling ranking as the criteria to reduce the dimension to the order of magnitude of the sample size, followed by machine learning using the decision tree, rule induction, random forest, and generalized linear model.

RESULTS

These four Machine learning models identified 66 potential HD-contributing genes, with the cross-validated accuracy of 90.79 ± 4.57%, 89.49 ± 5.20%, 90.45 ± 4.24%, and 97.46 ± 3.26%, respectively. The identified genes enriched the gene ontology of transcriptional regulation, inflammatory response, neuron projection, and the cytoskeleton. Moreover, three genes in the cognitive, sensory, and perceptual systems were also identified.

CONCLUSIONS

The mutant HTT may interfere with both the expression and transport of these identified genes to promote the HD pathogenesis.

背景

亨廷顿病（HD）是一种遗传性疾病，由 HTT 基因的多聚谷氨酰胺（poly-Q）突变引起，导致以舞蹈病、协调丧失、认知能力下降为特征的神经退行性变。然而，HD 的发病机制仍不清楚。尽管有广泛的生物数据可用，但迄今为止，尚未从机器学习的角度全面了解 HD 的发病机制，主要是由于缺乏所需的数据密度。

方法

为了从 157 例 HD 和 157 例对照的尸检前额叶皮层样本的表达谱中获取 HD 发病机制的知识，我们使用基因谱排序作为标准，将维度降低到样本数量的数量级，然后使用决策树、规则归纳、随机森林和广义线性模型进行机器学习。

结果

这四种机器学习模型确定了 66 个潜在的 HD 致病基因，交叉验证的准确率分别为 90.79±4.57%、89.49±5.20%、90.45±4.24%和 97.46±3.26%。鉴定出的基因丰富了转录调控、炎症反应、神经元投射和细胞骨架的基因本体。此外，还鉴定出了认知、感觉和知觉系统中的三个基因。