CAREER: Annotating the Microbiome using Machine Learning Methods

该项目解决了为复杂的现实世界问题开发复杂和新颖的机器学习技术的重要挑战。新技术使我们能够确定从海洋、土壤到人体的各种生态系统中共存的生物体的基因组。一些研究人员已经开始研究微生物组（定义为人体内微生物有机体的集合）对人类健康和疾病状况所发挥的致病作用。该CAREER项目的研究活动将开发用于从集体基因组样本中鉴定分类，功能和代谢潜力的方法。 一个关键的贡献将是多任务学习方法的发展，联合收割机信息跨多个层次的数据库与注释问题。在研究过程中，PI将调查捕获不同注释数据库中普遍存在的底层层次结构的最佳方法。这项研究的基本原理是，在几个人工管理的生物数据库中存在大量互补信息。 将微生物组与表型相关联需要整合各种高通量组学数据源（基因组学、代谢、蛋白质组学），这些数据源可能无法在所有样本中统一获得。 PI将在多任务学习范式中开发数据融合分类器，以整合异构、不完整的数据源来预测表型。该项目将作出以下主要贡献： (i)通过整合多个预测任务和相关数据库改进宏基因组注释模型。 (ii)在规则化多任务学习中引入层次信息。(iii)整合各种不完整的信息来源。(iv)可扩展的算法，使用基于哈希的特征表示并提高学习率。该项目是跨学科的，跨越机器学习，生物信息学，宏基因组学，微生物学和环境生态学领域。 这个项目将通过为所有学生提供一个智力和专业发展的环境，促进教学和研究之间的协同作用。 该项目将研究与教育计划相结合，重点是指导高中，本科和研究生，课程开发和实验室参观。 计划的活动包括培训跨学科研究人员、将微生物组分析相关项目整合到课堂中、课程改进和实施新的学习策略。作为该项目的一部分，将开发开源软件和工具，这将增强广泛多样的研究人员群体的科学理解和发现。

This project addresses an important challenge of developing sophisticated and novel machine learning techniques for complex real-world problems. New technologies allow us to determine the genomes of organisms co-existing within various ecosystems ranging from ocean, soil and human-body. Several researchers have embarked on studying the pathogenic role played by the microbiome, defined as the collection of microbial organisms within the human body, with respect to human health and disease conditions. The research activities in this CAREER project will develop approaches for the identification of taxonomy, function and metabolic potential from the collective genomes samples. A key contribution will be the development of multi-task learning approaches that combine information across multiple hierarchical databases associated with the annotation problems. During research, the PI will investigate the best ways to capture the underlying hierarchical structure, prevalent within different annotation databases. The rationale underlying this proposed research is that there is a wealth of complementary information that exists across several manually curated biological databases. Associating microbiome with phenotype requires integration of various high-throughput omic data sources (genomic, metabolic, proteomic) that may not be uniformly available across all samples. The PI will develop data fusion classifiers within the multi-task learning paradigm to integrate heterogeneous, incomplete data sources for predicting phenotypes. This project will lead to the following key contributions: (i) Improved metagenome annotation models by integration of multiple prediction tasks and associated databases. (ii) Incorporation of hierarchical information within regularized multi-task learning. (iii) Integration of diverse and incomplete information sources. (iv) Scalable algorithms that use hash based feature representations and improve the learning rates.This project is interdisciplinary and spans the fields of machine learning, bioinformatics, metagenomics, microbiology and environmental ecology. This project will foster the the synergy between teaching and research by providing an environment for all students to develop intellectually and professionally. The project integrates the research with an education plan focused on mentoring of high school, undergraduate and graduate students, curriculum development and laboratory visits. Planned activities include training of inter-disciplinary researchers, integration of microbiome analysis related projects within the classes, curriculum enhancement and implementation of new learning strategies. Open source software and tools will be developed as part of this project, that will enhance scientific understanding and discovery amongst a broad and diverse group of researchers.