Machine Learning Analysis of Genetic Modulators of Vaccine Immune Response

摘要 疫苗免疫反应遗传调节因子的机器学习分析。 这项建议描述了一种机器学习策略的发展，以识别相互作用的敏感性 多基因生物终点的基因座，重点是天花和炭疽疫苗相关的不良事件 (AES)和血清抗体反应的变异。接种天花疫苗后出现的不良反应 源于对炎症途径的过度刺激，可能受到多种相互作用的遗传因素的影响 各种因素。其中一些基因-基因相互作用可能是上位性的，对任何 单一变种。需要分析方法来测试全基因组数据中的关联，以说明 遗传变异之间的条件依赖关系，同时仍考虑具有高 边际效应。我们介绍了一种机器学习特征选择和优化方法，称为 蒸发冷却是一种以信息论和统计热力学为基础的冷却技术。 通过蒸发使粒子相互作用的系统。欧共体学习者的目标是识别 全基因组DNA序列数据中的易感基因或保护性基因。这种新颖的过滤方法， 不包括关于基因相互作用架构或相互作用顺序的假设，已被证明 确定一系列疾病易感性模型，包括边际主效应和纯交互作用 效果。在全基因组序列数据中表征多因素表型的遗传基础也是 由于存在大量的噪声变量，或者是 与表型无关。因此，EC算法蒸发(即，移除)噪声变量，留下 与给定表型相关的丰富变异的最小集合。我们建议将这一点提前 描述和解释单基因、基因-基因和基因-环境相互作用的方法 可能调节复杂的表型，如疫苗相关的AEs和人类免疫反应。这 策略将在人工数据的帮助下开发，在实际观察到的各种条件下进行模拟 数据，该策略将在单核苷酸多态(SNP)和志愿者的临床数据上进行测试 来自NIAID/NIH赞助的评估安万特巴斯德天花疫苗和疾病中心的试验 对照赞助试验评价炭疽疫苗吸附效果。

Abstract Machine Learning Analysis of Genetic Modulators of Vaccine Immune Response. This proposal describes the development of a machine-learning strategy to identify interacting susceptibility loci in polygenic biological endpoints, with a focus on smallpox and anthrax vaccine-related adverse events (AEs) and variation in serologic antibody response. The appearance of AEs following smallpox vaccination stems from excess stimulation of inflammatory pathways and is likely affected by multiple, interacting genetic factors. Some of these gene-gene interactions may be epistatic, having no distinct marginal effect for any single variant. Analytical approaches are needed for testing association in genome-wide data to account for conditional dependencies between genetic variants while still accounting for co-occurring variants with high marginal effects. We have introduced a machine-learning feature selection and optimization method called Evaporative Cooling (EC), which is based on information theory and the statistical thermodynamics of cooling a system of interacting particles by evaporation. The objective of the EC learner is the identification of susceptibility or protective genes in genome-wide DNA sequence data. This novel filter method, which includes no assumptions regarding gene interaction architecture or interaction order, has been shown to identify a spectrum of disease susceptibility models, including marginal main effects and pure interaction effects. Characterizing the genetic basis of multifactorial phenotypes in genome-wide sequence data is also computationally challenging due to the presence of a large number of noise variants, or variants that are irrelevant to the phenotype. Thus, the EC algorithm evaporates (i.e., removes) noise variants, leaving behind a minimal collection of variants enriched for relevance to the given phenotype. We propose to advance this method to characterize and interpret singe-gene, gene-gene and gene-environment interactions all of which may modulate complex phenotypes such as vaccine-associated AEs and human immune response. This strategy will be developed with the aid of artificial data, simulated under a variety of conditions observed in real data, and the strategy will be tested on single nucleotide polymorphism (SNP) and clinical data from volunteers from a NIAID/NIH-sponsored trial to evaluate the Aventis Pasteur Smallpox Vaccine and a Center for Disease Control sponsored trial to evaluate Anthrax Vaccine Adsorbed.