Algorithm-based prevention and reduction of cancer health disparity arising from data inequality

少数民族在生物医学研究和临床方面长期处于积累的数据劣势 学习。统计数据显示，在与癌症相关的GWA和临床组学项目中，超过90%的样本 是从欧洲血统的人那里收集的。少数民族的这种严重的数据劣势 随着数据驱动的、基于算法的生物医学研究和临床，组织将产生新的健康差距 决策变得越来越普遍。由数据不平等引起的新的癌症差异可能会 在存在数据不平等的所有类型的癌症中影响所有少数民族群体。因此，它的负面影响是 不限于已经明显存在明显种族差异的癌症类型或亚型。这个 拟议研究的长期目标是防止或减少数据引起的健康差异 少数民族的劣势。这项工作的总体目标是获取关键知识并创造 开放资源，利用多种族临床组学数据建立机器学习的新范式。我们的中央 假设从大多数人的数据中学习到的知识可以被转移到改进 机器学习在数据弱势少数民族群体上的表现。以强劲的前期工作为指导 数据，我们将追求两个具体目标：1)从癌症临床组学数据中发现和基因-表型 数据：在什么条件下，迁移学习方案在多大程度上改进了机器学习模型 对数据弱势少数民族群体的绩效；2)为无偏见的 多民族机器学习，防止或减少因数据劣势而产生的新的健康差距 少数民族。这种方法是创新的，因为它代表着对现状的实质性偏离 通过将多民族机器学习的范式从混合学习和自主学习转变为 方案转换为迁移学习方案。这项拟议的研究意义重大，因为它有望确定 无偏见多民族机器学习的新范式，并提供开放的资源系统，以促进 范式转变，从而防止或减少因族裔数据劣势而产生的健康差距 少数族裔。

Ethnic minority groups have a long-term cumulative data disadvantage in biomedical research and clinical studies. Statistics have shown that over 90% of the samples in cancer-related GWAS and clinical omics projects were collected from Individuals of European ancestry. This severe data disadvantage of the ethnic minority groups is set to produce new health disparities as data-driven, algorithm-based biomedical research and clinical decisions become increasingly common. The new cancer disparity arising from data inequality can potentially impact all ethnic minority groups in all types of cancers where data inequality exists. Thus, its negative impact is not limited to the cancer types or subtypes for which significant ethnic disparities have already been evident. The long-term goal of the proposed research is to prevent or reduce the heath disparities arising from the data disadvantage of ethnic minority groups. The overall objective of this work is to obtain key knowledge and create open resources to establish a new paradigm for machine learning with multiethnic clinical omics data. Our central hypothesis is that the knowledge learned from data of the majority population can be transferred to improve machine learning performance on the data-disadvantaged ethnic minority groups. Guided by strong preliminary data, we will pursuit two specific aims to 1) Discover from cancer clinical omics data and genotype-phenotype data: under what conditions and to what extent the transfer learning scheme improves machine learning model performance on data-disadvantaged ethnic minority groups; 2) Create an open resource system for unbiased multiethnic machine learning to prevent or reduce new health disparities arising from the data disadvantage of ethnic minorities. The approach is innovative because it represents a substantive departure from the status quo by shifting the paradigm of multiethnic machine learning from mixture learning and independent learning schemes to a transfer learning scheme. The proposed research is significant, because it is expected to establish a new paradigm for unbiased multiethnic machine learning and to provide an open resource system to facilitate the paradigm shift, and thus to prevent or reduce health disparities arising from the data disadvantage of ethnic minorities.