Peripheral blood MicroRNAs as biomarkers of schizophrenia: expectations from a meta-analysis that combines deep learning methods

This study aimed at identifying reliable differentially expressed miRNAs (DEMs) for schizophrenia in blood <i>via</i> meta-analyses combined with deep learning methods. First, we meta-analysed published DEMs. Then, we enriched the pool of schizophrenia-associated miRNAs by applying two computational learning methods to identify candidate biomarkers and verified the results in external datasets. In total, 27 DEMs were found to be statistically significant (<i>p</i> &lt; .05). Ten candidate schizophrenia-associated miRNAs were identified through computational learning methods. The diagnostic efficiency was verified on a blood-miRNA dataset (GSE54578) with a random forest (RF) model and achieved an area under the curve (AUC) of 0.83 ± 0.14. Moreover, 855 experimentally validated target genes for these candidate miRNAs were retrieved, and 11 hub genes were identified. Enrichment analysis revealed that the main functions in which the target genes were enriched were those related to cell signalling, prenatal infections, cancers, cell deaths, oxidative stress, endocrine disorders, transcription regulation, and kinase activities. The diagnostic ability of the hub genes was reflected in a comparably good average AUC of 0.77 ± 0.09 for an external dataset (GSE38484). A meta-analysis that combines computational and mathematical methods provides a reliable tool for identifying candidate biomarkers of schizophrenia.

本研究旨在通过结合荟萃分析（meta-analyses）与深度学习方法（deep learning methods），筛选出可用于精神分裂症诊断的血液样本差异表达微小RNA（differentially expressed miRNAs，DEMs）。首先，我们对已发表的DEMs进行了荟萃分析。随后，我们通过两种计算学习方法（computational learning methods）筛选候选生物标志物（biomarkers），以扩充精神分裂症相关微小RNA库，并在外部数据集上验证了所得结果。最终共筛选出27个具有统计学意义的DEMs（p < 0.05）。通过计算学习方法，我们共鉴定出10个精神分裂症相关候选微小RNA。我们在血液微小RNA数据集GSE54578上采用随机森林（random forest，RF）模型验证了诊断效能，所得受试者工作特征曲线下面积（area under the curve，AUC）为0.83 ± 0.14。此外，我们检索得到这些候选微小RNA的855个经实验验证的靶基因，并从中鉴定出11个枢纽基因（hub genes）。富集分析结果显示，靶基因主要富集于与细胞信号传导（cell signalling）、产前感染、癌症、细胞死亡、氧化应激（oxidative stress）、内分泌紊乱（endocrine disorders）、转录调控（transcription regulation）及激酶活性（kinase activities）相关的生物学功能中。在外部数据集GSE38484中，枢纽基因的诊断能力同样表现优异，平均AUC达0.77 ± 0.09。本研究采用的结合计算与数学方法的荟萃分析，为精神分裂症候选生物标志物的筛选提供了可靠的研究工具。