机器学习分类器揭示梅尼埃病基因组编码区中隐藏的罕见双等位基因杂合变异

文献信息

中文摘要

背景： 梅尼埃病（MD）是一种罕见的内耳疾病，具有显著的遗传贡献。我们开发了MD分类器，一种机器学习工具，利用编码区的罕见变异来区分家族性和散发性MD病例。我们构建了一个疾病特异性基因变异评分，整合了基因、变异和疾病特异性特征。梯度提升分类器取得了最高性能，测试AUC为0.843（95% CI，0.702-0.948），准确率为81.5%。将散发病例重新分类为家族性组的比例为16.6%至28.3%。该模型识别出双等位基因杂合变异的负荷，提示可能存在隐性遗传，可能影响高达38%的MD患者。使用主要分类器特征进行的基因富集分析揭示了家族性和散发性MD之间不同的生物学通路。这一基于机器学习的框架整合了独特的遗传特征，可能提高罕见疾病的诊断率，并普遍适用于同时具有家族性和散发性表型的疾病，从而能够识别复合杂合隐性遗传。

英文摘要

Meniere disease (MD) is a rare inner ear disorder with a significant genetic contribution. We developed the MD classifier, an ML tool to distinguish familial from sporadic MD cases using rare variants in coding regions. We have generated a disease-specific gene variant score incorporating gene, variant, and disease-specific features. A gradient boosting classifier achieved the highest performance, with a test AUC of 0.843 (95% CI, 0.702-0.948) and an accuracy of 81.5%. Reclassification of sporadic cases into the familial group ranged from 16.6% to 28.3%. The model identified a burden of biallelic heterozygous variants suggesting hidden recessive inheritance that may affect up to 38% of MD patients. Gene-enrichment analysis using major classifier features revealed distinct biological pathways between familial and sporadic MD. This ML-based framework integrates unique genetic features that may improve the diagnostic yield of rare disorders and is universally applicable to conditions with both familial and sporadic phenotypes, enabling the identification of compound heterozygous recessive inheritance.