School of Computer Science, McGill University, Rue University, Montréal, Quebec, Canada.
Department of Psychiatry, McGill Group for Suicide Studies, Douglas Mental Health University, Montreal, Quebec, Canada.
PLoS One. 2023 May 15;18(5):e0285123. doi: 10.1371/journal.pone.0285123. eCollection 2023.
Major depressive disorder (MDD) is a leading cause of disability worldwide, and is commonly treated with antidepressant drugs (AD). Although effective, many patients fail to respond to AD treatment, and accordingly identifying factors that can predict AD response would greatly improve treatment outcomes. In this study, we developed a machine learning tool to integrate multi-omic datasets (gene expression, DNA methylation, and genotyping) to identify biomarker profiles associated with AD response in a cohort of individuals with MDD.
Individuals with MDD (N = 111) were treated for 8 weeks with antidepressants and were separated into responders and non-responders based on the Montgomery-Åsberg Depression Rating Scale (MADRS). Using peripheral blood samples, we performed RNA-sequencing, assessed DNA methylation using the Illumina EPIC array, and performed genotyping using the Illumina PsychArray. To address this rich multi-omic dataset with high dimensional features, we developed integrative Geneset-Embedded non-negative Matrix factorization (iGEM), a non-negative matrix factorization (NMF) based model, supplemented with auxiliary information regarding gene sets and gene-methylation relationships. In particular, we factorize the subjects by features (i.e., gene expression or DNA methylation) into subjects-by-factors and factors-by-features. We define the factors as the meta-phenotypes as they represent integrated composite scores of the molecular measurements for each subject.
Using our model, we identified a number of meta-phenotypes which were related to AD response. By integrating geneset information into the model, we were able to relate these meta-phenotypes to biological processes, including a meta-phenotype related to immune and inflammatory functions as well as other genes related to depression or AD response. The meta-phenotype identified several genes including immune interleukin 1 receptor like 1 (IL1RL1) and interleukin 5 receptor (IL5) subunit alpha (IL5RA), AKT/PIK3 pathway related phosphoinositide-3-kinase regulatory subunit 6 (PIK3R6), and sphingomyelin phosphodiesterase 3 (SMPD3), which has been identified as a target of AD treatment.
The derived meta-phenotypes and associated biological functions represent both biomarkers to predict response, as well as potential new treatment targets. Our method is applicable to other diseases with multi-omic data, and the software is open source and available on Github (https://github.com/li-lab-mcgill/iGEM).
重度抑郁症（MDD）是全球导致残疾的主要原因，通常采用抗抑郁药物（AD）进行治疗。尽管有效，但许多患者对 AD 治疗没有反应，因此确定可以预测 AD 反应的因素将极大地改善治疗结果。在这项研究中，我们开发了一种机器学习工具，将多组学数据集（基因表达、DNA 甲基化和基因分型）整合在一起，以识别与 MDD 患者队列中 AD 反应相关的生物标志物特征。
111 名 MDD 患者接受 8 周的抗抑郁药治疗，并根据蒙哥马利-阿斯伯格抑郁评定量表（MADRS）将他们分为反应者和无反应者。使用外周血样本，我们进行了 RNA 测序，使用 Illumina EPIC 阵列评估 DNA 甲基化，并使用 Illumina PsychArray 进行基因分型。为了解决这个具有高维特征的丰富多组学数据集，我们开发了整合基因集嵌入式非负矩阵分解（iGEM），这是一种基于非负矩阵分解（NMF）的模型，补充了有关基因集和基因-甲基化关系的辅助信息。特别是，我们通过特征（即基因表达或 DNA 甲基化）将受试者分解为受试者-因子和因子-特征。我们将这些因子定义为元表型，因为它们代表了每个受试者的分子测量的综合综合评分。
使用我们的模型，我们确定了一些与 AD 反应相关的元表型。通过将基因集信息整合到模型中，我们能够将这些元表型与生物学过程相关联，包括与免疫和炎症功能相关的元表型以及其他与抑郁或 AD 反应相关的基因。该元表型鉴定了几个基因，包括免疫白细胞介素 1 受体样 1（IL1RL1）和白细胞介素 5 受体（IL5）亚基α（IL5RA）、AKT/PI3 途径相关的磷酸肌醇-3-激酶调节亚基 6（PIK3R6）和鞘磷脂磷酸二酯酶 3（SMPD3），它已被确定为 AD 治疗的靶点。
推导的元表型和相关的生物学功能不仅代表了预测反应的生物标志物，还代表了潜在的新治疗靶点。我们的方法适用于具有多组学数据的其他疾病，并且该软件是开源的，并可在 Github（https://github.com/li-lab-mcgill/iGEM）上获得。
