Machine learning of RNA code artifacts in inflammatory diseases towards precision medicine

背景：RNA核苷酸的内源性修饰（“表转录组”）最近成为影响几乎所有生物过程和疾病的RNA命运的关键后/共转录决定因素。目前尚不清楚驱动多种炎症性疾病（如动脉粥样硬化性心脏病和自身免疫性疾病）的常见分子决定因素/变化是什么。我们之前已经在慢性炎症性疾病患者中建立了催化最普遍的RNA修饰的酶的惊人调节。我们现在希望利用创新的表转录组学（跨基因组研究表转录组的方法）和机器学习方法来利用这些知识，将我们的实验发现应用于个性化医疗。在此，我们的目标是：1)在健康个体和慢性炎症性疾病患者的转录组（=表转录组）上绘制两种最常见的RNA修饰（腺苷脱氨和m6A RNA甲基化）；2)确定表转录组的差异调控及其对疾病期间基因表达和生物学过程的影响；3)将差异表转录组动力学（位点和基因表达变化）与临床数据结合机器学习方法，以确定RNA修饰在慢性炎症性疾病中的预后和抗炎药物反应预测价值。新颖性和及时性：随着新一代测序技术的进步和基于抗体的RNA修饰识别方法的发展，对腺苷RNA修饰的深入研究现在是可以实现的。然而，如果没有解释方法，基因组学和临床数据的结合具有有限的效用。过去几年出现的机器学习已被证明能够可靠地识别临床相关模式。通过采用这些前沿方法，我们将首次合成炎症性疾病的表转录谱，并将提供与患者分层、药物反应和临床结果预测功能相关的新分子决定因素（例如特定位点的修饰）。实验方法:在目标1和2中，我们将绘制和差异量化自身和可用的散装和甲基化RNA-seq数据中的腺苷RNA修饰，这些数据来自高表型患者的外周血和其他相关组织（心脏、滑膜），这些患者患有类风湿关节炎、狼疮、系统性硬化症、动脉粥样硬化性心脏病和健康人通过主要导师和第三导师实验室建立的管道。在目标3中，与Second-supervisor一起，我们将遵循监督机器学习方法，通过使用我们的数据训练算法，测试这些表转录组学变化是否预测疾病程度、进展和对当前治疗方案的反应，在大型可用数据集中具有显著的净重分类指数（>20%），超过当前基于临床的算法。

Background: Endogenous modifications of RNA nucleotides ("epitranscriptome") have recently emerged as critical post/co-transcriptional determinants of RNA fate affecting almost every biological process and disease. Which are the common molecular determinants/changes driving multiple inflammatory diseases, like atherosclerotic heart disease and autoimmune disorders, are currently unclear. We have previously established a striking regulation of enzymes catalyzing the most prevalent RNA modifications in patients with chronic inflammatory disorders. We now want to harness this knowledge using innovative epitranscriptomics (methods studying epitranscriptome across genome) and machine learning approaches to apply our bench findings in personalized medicine. Objectives: Herein, we aim to: 1) map the two most prevalent RNA modifications (adenosine deamination and m6A RNA methylation) on the transcriptome (=epitranscriptome) of healthy individuals and patients with chronic inflammatory disorders, 2) define the differential regulation of epitranscriptome and its effect on gene expression and biological processes during disease, 3) integrate the differential epitranscriptome dynamics (sites and gene expression changes) combined with clinical data in machine learning approaches in order to determine the prognostic and anti-inflammatory drug-response predictive value of RNA modifications in chronic inflammatory diseases. Novelty and timeliness: With the recent advancement of next-generation sequencing and development of antibody-based methods that recognize RNA modifications, an in-depth study of adenosine RNA modifications is now accessible. However, combination of 'omic and clinical data bears restricted utility without interpretation methods. Machine learning, emerging the last couple of years, has been proven to reliably identify clinically relevant patterns. By employing these cutting-edge methods, we will synthesize for the first time the epitranscriptional profile of inflammatory diseases and will provide novel molecular determinants (e.g modification of specific sites) functionally relevant to stratification of patients, drug-responses and prediction of clinical outcomes. Experimental approach: In Objective 1 and 2, we will map and differentially quantify the adenosine RNA modifications in own and available bulk and methylated RNA-seq data from peripheral blood and other relevant tissues (cardiac, synovial) from highly phenotyped patients (for disease activity and anti-inflammatory drug-response where applicable) with rheumatoid arthritis, lupus, systemic sclerosis, atherosclerotic heart disease and in healthy individuals through established pipelines at the lab of Main-supervisor and Third-supervisor. In Objective 3, together with Second-supervisor, we will follow a supervised machine learning approach, by training algorithms using our data, to test whether these epitranscriptomic changes predict disease extent, progression and response to current therapeutic options in large available datasets with significant net reclassification index (>20%) over current clinical-based algorithms.