Machine Learning and Exosome Derived Biomarkers of Obstructive Sleep Apnea Induced Hypertension

点击翻译按钮获取中文摘要

Obstructive sleep apnea (OSA) can lead to hypertension via chronic intermittent hypoxia and sleep fragmentation. There are no biomarkers of OSA induced hypertension to target blood pressure (BP) control with OSA treatments and prevent cardiovascular disease. The objective of this proposal is to develop prognostic biomarkers of OSA induced hypertension. We will conduct a prospective case-control study in patients with newly diagnosed moderate to severe OSA (cases) or no OSA (controls) to achieve this objective. We will define OSA induced hypertension by characteristically abnormal BP profiles during sleep: post-apnea acute surges in BP, non-dipping BP, and nocturnal hypertension. Our proof-of-concept study used machine learning methods to devise a novel type of Long-Short Term Memory (LSTM) Encoder- Decoder network to predict BP during sleep using physiological signals from polysomnography. The first aim of this proposal is to train and validate an LSTM network to predict BP profiles during sleep from clinical polysomnography signals compared to non-invasive beat-to-beat BP measurements (Finapres Nova) as reference. An LSTM network that predicts nocturnal BP profiles from polysomnography will provide a clinically applicable digital biomarker for OSA induced hypertension. Our pilot study showed that a small cluster of exosome micro RNAs (miRNA), implicated in cardiovascular disease, are differentially expressed in OSA patients with non-dipping BP. Exosome cargo is specifically sorted, expressed stably in individuals, and functions as an essential vehicle of targeted inter-cellular genomic communication. We will build on this study with the second aim to identify genomic biomarkers of OSA induced hypertension in exosome cargo by miRNA and mRNA sequencing, proteomics, and lipidomics, followed by bioinformatics for differential expression between cases and controls. The small studies examining physiological pathways that mediate OSA induced hypertension have not systematically included multiple dysregulated pathways: sympathetic and renin-angiotensin-aldosterone activation, oxidative stress, and inflammation. Our third aim is to develop a parsimonious and robust biomarker panel by combining molecular assays of physiological dysregulation (blood renin, angiotensin II, aldosterone, high sensitivity C-reactive protein, tumor necrosis factor-α, malondialdehyde, and urine catecholamines and isoprostane) and exosome- derived omics biomarkers with a retrained LSTM network. The optimal biomarker panel will be determined by network performance and clinical applicability. This proposal includes a multidisciplinary team of investigators with expertise in clinical OSA research, machine learning, exosome biology and omics, biostatistics, and bioinformatics. This study will have a high clinical impact by providing accessible prognostic biomarkers of OSA induced hypertension to guide pivotal management decisions and future interventional research in OSA to reduce cardiovascular disease.