Machine Learning and Multiomics for Predictive Models and Biomarker Discovery in Preterm Infants.

项目摘要 出生时< 32周且<1500 g的早产儿（极低出生体重，VLBW）死亡率增加 （10-15%）和不到70%的生存没有重大发病率。微生物生态失调与 主要早产发病率，但微生物代谢产物或其影响机制 病理生理学、存活率和发病率尚不清楚。该提案的目的是发展全面的 整合临床数据和多组学特征的预测模型，有助于生物标志物的发现， 从传统到靶向精准医疗的新生儿医学范式。总体假设是 将代谢和多组学特征与临床数据相结合将可靠地预测生存率和主要风险。 早产儿、极低出生体重儿的发病率。本研究的长期目标是建立因果关系 微生物代谢产物与早产儿疾病之间的关系，有助于了解 微生物代谢产物和改善早产结局。我们将使用以下特定的方法来测试我们的假设 1）利用机器学习技术开发死亡率和死亡率的临床预测模型， 早产儿、极低出生体重儿的特定发病率：我们将检验这一假设，即一个整合临床 前2周龄的变量，将准确预测死亡率和晚发性脓毒症，NEC， BPD、重度ROP和重度IVH。我们将采用佛蒙特州牛津数据库的回顾性队列 (VON)得克萨斯州儿童医院，（n= 3385 VLBW婴儿）。我们将验证临床预测模型 来自目标1A，前2周的前瞻性临床数据，来自目标2（n=300），目标2） 描述微生物代谢产物和多组学特征，区分早产VLBW婴儿与 死亡率和发病率，完善预测模型，加强生物标志物的发现：我们将测试假设 使用机器学习技术将多组学特征与临床数据相结合， 预测模型（死亡率和迟发性脓毒症的特异性发病率、NEC、BPD、ROP和IVH/PVL） 更好的准确性和增强生物标志物的发现。我们将在一项前瞻性研究设计中实现这一点， 入组早产儿（<32周）、极低出生体重儿（n= 300），纵向收集粪便、尿液和血液样本 每周两次，持续2周龄。我们期望识别已知和新的代谢物， 代谢途径迄今尚未确定的影响早产儿的病理生理和结果。整体 使用生命前两周的信息建立预测模型，将使我们能够及早引入干预措施。 改善健康轨迹和患者结果，从而促进主动精确的范例 新生儿医学我们的研究结果的影响超出了泌尿外科学领域，扩展到其他患者 以及微生物生态失调和代谢组改变是发病机制中的关键因素的疾病。

PROJECT SUMMARY Preterm infants born at < 32 weeks and <1500 g (very low birth weight, VLBW) suffer from increased mortality (10-15%) and less than 70% survive without major morbidity. Microbial dysbiosis has been associated with major preterm morbidities but the microbial metabolites or the mechanisms by which they impact pathophysiology, survival and morbidity is not known. The purpose of this proposal is to develop holistic prediction models integrating clinical data and multi-omic signatures, aid biomarker discovery and advance the paradigm in Neonatal Medicine from traditional to targeted precision medicine. The overarching hypothesis is that integrating metabolic and multi-omic signatures with clinical data will reliably predict survival and major morbidity in preterm, VLBW infants. The long-term goal of this research is to establish causal association between identified microbial metabolites and disease in preterm infants, contribute to the knowledgebase of microbial metabolites and improve preterm outcomes. We will test our hypothesis using the following Specific Aims; Aim 1) Leverage machine learning techniques to develop clinical prediction models for mortality and specific morbidities in preterm, VLBW infants: We will test the hypothesis, that a model integrating clinical variables in the first 2 wks. of age, will accurately predict mortality, and morbidities of late-onset sepsis, NEC, BPD, severe ROP and severe IVH. We will employ a retrospective cohort from the Vermont Oxford Database (VON) from Texas Children’s Hospital, (n= 3385 VLBW infants). We will validate the clinical predictive models derived from aim 1A with the prospective clinical data from the first 2 weeks, from Aim 2 (n=300), Aim 2) Delineate microbial metabolites and multi-omic signatures that differentiate preterm VLBW infants with mortality and morbidity, refine predictive models and enhance biomarker discovery: We will test the hypothesis that integrating multi-omics signatures with clinical data using machine learning techniques will refine our predictive models (mortality and specific morbidities of late-onset sepsis, NEC, BPD, ROP and IVH/PVL) for better accuracy and enhance biomarker discovery. We will accomplish this in a prospective study design of enrolled preterm (< 32weeks), VLBW infants (n= 300) and collect stool, urine and blood samples, longitudinally twice a week for 2 weeks of age. We anticipate identifying known and novel metabolites and delineating metabolic pathways hitherto unidentified that influence preterm pathophysiology and outcomes. Holistic prediction models using information from the first 2 weeks of life will enable us to introduce interventions early to improve health trajectories and patient outcomes, thereby facilitating the paradigm of proactive precision medicine in Neonatology. The impact of our results extend beyond the field of neonatology, to other patients and diseases where microbial dysbiosis and altered metabolome are key factors in the pathogenesis.