III: Small: Collaborative Research: High-Dimensional Machine Learning Methods for Personalized Cancer Genomics

个性化和精确癌症基因组学成功的关键在于：(1)发现和理解遗传改变如何影响与癌症相关的各种细胞过程的分子水平机制；(2)利用分子特征为患者量身定制更个性化的治疗策略。为了实现这些目标，近年来发展了各种高通量实验方法，以获取患者肿瘤基因组序列、mRNA表达、蛋白表达、表观遗传读出等患者肿瘤的详细信息。然而，充分利用如此大量的高维数据来产生生物医学见解的算法往往是缺乏的。该项目将通过开发使用高维机器学习新兴技术的新算法，推进数据驱动的癌症基因组数据复杂建模领域，以实现个性化癌症治疗。这项研究的结果有可能影响机器学习领域和计算基因组学领域。与研究项目相结合的教育部分将开发新的课程材料，让本科生和未被充分代表的群体参与研究，并培养新一代跨学科研究生研究人员。该项目由两个协同研究重点组成，旨在开发用于分析高通量癌症基因组数据的新型高维机器学习算法。首先，该项目将开发用于多视图数据建模的高维图形模型，以整合来自异构全基因组数据源的数据。第二，为个性化药物推荐设计新颖的高维协同学习方法。高维图形模型将用于估计不同癌症亚型的网络。然后将这些网络集成到推荐算法中，从而有助于改进多视图图形模型估计。该项目将通过精确定位复杂分子相互作用在癌症发生和发展中的作用，增强对大规模癌症基因组数据的解释能力，这将为更有效地发现个性化分子特征和更有针对性的癌症潜在治疗提供新的方法。这种技术创新和概念进步有可能重塑人们处理图形模型估计及其在生物环境中的作用的方式。该项目将潜在地为机器学习和计算生物学以及其他学科的理论家和实践者开辟新的可能性。

The key to success in personalized and precision cancer genomics lies in: (1) discovering and understanding the molecular-level mechanisms of how genetic alterations influence various cellular processes relevant to cancer, and (2) utilizing molecular signatures to tailor more personalized treatment strategies for patients. In order to achieve these goals, various high-throughput experimental methods have been developed in recent years to obtain information about a patient's cancer genome sequence, mRNA expression, protein expression, epigenetic readout, and other detailed information about a patient's tumor. However, algorithms that fully harness such a massive amount of high dimensional data to yield biomedical insights are often lacking. This project will advance the field of data-driven complex modeling of cancer genomic data for personalized cancer treatment by developing novel algorithms that use emerging and new techniques in high-dimensional machine learning. The results of this research have the potential to impact both the machine learning field and the computational genomics field. The educational components integrated with the research program will develop new curriculum materials, involve undergraduate students and underrepresented groups in research, and train a new generation of interdisciplinary graduate researchers. This project consists of two synergistic research thrusts to develop novel high-dimensional machine learning algorithms for analyzing high-throughput cancer genomic data. First, the project will develop high-dimensional graphical models for multi-view data modeling to integrate data from heterogeneous genome-wide data sources. Second, it will devise novel high-dimensional collaborative learning methods for personalized drug recommendation. The high-dimensional graphical models will be used to estimate networks for different cancer subtypes. These networks will then be integrated into the recommendation algorithms, which in turn will help improve the multi-view graphical model estimation. This project will enhance the ability to interpret large-scale cancer genomics data by pinpointing the roles of complex molecular interactions in cancer onset and progression, which will enable novel ways to more effectively discover personalized molecular signatures and more targeted potential treatments of cancer. Such technical innovation and conceptual advancement have the potential to reshape the way that one approaches graphical model estimation and its role in biological contexts. The project will potentially open up new possibilities for both theoreticians and practitioners in machine learning and computational biology as well as other disciplines.