Robust, Generalizable, and Fair Machine Learning Models for Biomedicine

项目摘要 现代机器学习方法在以下方面取得了实质性的成功 模式识别和高维数据分析。然而，这些算法 严重依赖于关联发现，这无法阐明其机制 支持观察到的相关性，并遭受有限的概括性。至 为了应对这一挑战，Yu Lab专注于开发强大的 可推广的机器学习方法集成各种类型的生物医学 数据，包括多组学、病理学和表型信息。下一步的目标 五年的时间是开发新的计算方法，将机器学习与 带有因果推理方法的算法，以更好地理解分子 支持疾病病理学的机制，并使公平和可靠的预测成为可能 药物反应和毒性。拟议研究计划的总体愿景是 建立可推广的数据驱动方法，将生物医学数据转化为健壮的数据 预测模型和机械模型。拟议的方法将系统地将 不同的生物医学信号，以提取以前未知的分子知识 机制，并得出可靠的预测模型的药物的效果。这个 拟议的方法是创新的，因为它们通过以下方式改变了现状 将先进的因果推理技术与数据驱动算法相结合 增强机械性和预测性建模。这项研究计划具有重要意义 因为它有望提高我们对疾病病理学的理解，并提供一种 药物反应和不良反应的公平和可推广的信息学框架 在不同人群中的预测。拟议的研究活动将开启新的 通过建立一个新的机器学习平台来研究视界 可靠的预测，这将垂直推进分子生物学、药理学和 生物医学中的计算研究。

Project Summary Modern machine learning approaches have attained substantial success in pattern recognition and high-dimensional data analyses. However, these algorithms heavily rely on association discovery, which cannot elucidate the mechanisms underpinning the observed correlations and suffers from limited generalizability. To address this challenge, the Yu Lab focuses on the development of robust and generalizable machine learning approaches to integrate various types of biomedical data, including multi-omics, pathology, and phenotypic information. The goal of the next five years is to develop novel computational methods that connect machine learning algorithms with causal inference methodologies to better understand the molecular mechanisms underpinning disease pathology and enable fair and robust predictions of drug response and toxicity. The overall vision of the proposed research program is to establish generalizable data-driven methods to transform biomedical data into robust prediction and mechanistic models. The proposed approach will systematically connect diverse biomedical signals to extract previously unknown knowledge on the molecular mechanisms and derive reliable prediction models for the effects of medications. The proposed approaches are innovative because they depart from the status quo by incorporating advanced causal inference techniques with data-driven algorithms to enhance mechanistic and predictive modeling. This research program is significant because it is expected to improve our understanding of disease pathology and provide a fair and generalizable informatics framework for drug response and adverse effects prediction in diverse populations. The proposed research activities will open new research horizons by establishing a new machine learning platform for generating reliable predictions, which will vertically advance molecular biology, pharmacology, and computational research in biomedicine.