Previous studies have provided evidence for an alteration of genetic coexpression in schizophrenia (SCZ). However, such analyses have thus far lacked biological specificity for individual genes, which may be critical for identifying illness-relevant effects. Therefore, we applied machine learning to identify gene-specific coexpression differences at the individual subject level and compared these between individuals with SCZ, bipolar disorder, major depressive disorder (MDD), autism spectrum disorder (ASD), and healthy controls. Utilizing transcriptome-wide gene expression data from 21 independent datasets, comprising a total of 9509 participants, we identified a reproducible decrease of BCL11A coexpression across 4 SCZ datasets that showed diagnostic specificity for SCZ when compared with ASD and MDD. We further demonstrate that individual-level coexpression differences can be combined in multivariate coexpression scores that show reproducible illness classification across independent datasets in SCZ and ASD. This study demonstrates that machine learning can capture gene-specific coexpression differences at the individual subject level for SCZ and identify novel biomarker candidates.

中文翻译：

---

通过共表达的个体水平预测鉴定精神分裂症中可重复的 BCL11A 改变


先前的研究已经为精神分裂症（SCZ）基因共表达的改变提供了证据。然而，此类分析迄今为止缺乏单个基因的生物学特异性，这对于识别疾病相关影响可能至关重要。因此，我们应用机器学习来识别个体受试者水平上的基因特异性共表达差异，并比较 SCZ、双相情感障碍、重度抑郁症 (MDD)、自闭症谱系障碍 (ASD) 和健康对照个体之间的差异。利用来自 21 个独立数据集（总共包括 9509 名参与者）的全转录组基因表达数据，我们发现 4 个 SCZ 数据集中BCL11A共表达出现了可重复的下降，这表明与 ASD 和 MDD 相比，SCZ 具有诊断特异性。我们进一步证明，个体水平的共表达差异可以合并到多变量共表达评分中，这些评分显示 SCZ 和 ASD 独立数据集中可重复的疾病分类。这项研究表明，机器学习可以捕获 SCZ 在个体受试者水平上的基因特异性共表达差异，并识别新的候选生物标志物。