III: Small: Network Learning for Integrative Cancer Genomics

新的大规模DNA测序和阵列技术通过产生测量癌症基因组中异常的大量信息，为研究癌症的分子机制提供了一条很有前途的途径。基因组信息可以潜在地指导靶向分子的药物设计，并改善癌症治疗的临床决策。进一步发展的主要障碍之一是从庞大的基因组数据中阐明多种复杂的癌症分子标志物。这一建议解决了基于网络的机器学习理论框架和方法的问题，这些理论框架和方法可以为癌症基因组信息和相关生物医学知识的综合研究建立潜在的生物学机制模型。作为概念的验证，所开发的方法将用于研究卵巢癌治疗中的化疗耐药性。这项建议旨在创建一个通用的计算驱动的方法来指导癌症基因组学研究，并改进癌症治疗中基于基因组学的临床决策。提案中描述的研究活动将提供一系列有效和高效的计算工具，将异质基因组数据与生物医学知识结合起来用于临床实践。对卵巢癌数据的研究将有助于揭示导致化疗耐药的关键途径，并为卵巢癌的治疗提供有用的预测工具和药物靶点。这项提议还将把计算癌症基因组学的最新研究进展整合到几个培训计划的新课程中，为学生未来的职业做好准备，以满足中西部上游地区不断增长的生物医学和健康信息产业的劳动力需求。教育计划还将重点招收少数群体和代表不足的计算机科学和信息技术的学生。为了实现这些目标，研究计划的组成部分是：1)制定图形核和子图挖掘算法，可以整合各种类型的癌症基因组异常，以改善癌症结果预测，并发现致癌基因组异常模式；2)利用表型、基因及其关联网络中的已知关系，制定具有拉普拉斯约束的半监督矩阵因式分解方法，用于预测新的癌症表型和基因关联，以识别潜在的药物靶点；3)研究卵巢癌治疗中的化疗耐药，揭示卵巢癌耐药的关键途径，为卵巢癌的治疗开发有用的预测工具和药物靶点；4)将所开发的方法在软件包和网络工具中发布，供学术界使用。教育计划的两个主要组成部分是：1)在明尼苏达州BioSMART计划的暑期学院为明尼苏达州的高中生提供为期两周的名为“用计算机治疗癌症”的课程；2)创建一门新课程--生物医学信息学中的计算基因组学，以支持两个研究生计划，以培训具有基因组学和计算机科学知识的生物医学/健康信息学学生。

New large-scale DNA sequencing and array technologies now provide a promising way to study the molecular mechanisms of cancer by generating enormous information measuring aberrations in cancer genome. The genomic information can potentially guide drug design on targeted molecules, and improve clinical decisions in cancer treatment. One of the main obstacles to further progress is to elucidate multiple complex molecular indicators of cancers from the enormous genomic data. This proposal tackles the problem with network-based machine-learning theoretical frameworks and methods that can model the underlying biological mechanisms for an integrative study of cancer genomic information and relevant biomedical knowledge. As a proof of concept, the developed methods will be applied to study chemoresistance in ovarian cancer treatment. This proposal aims at creating a general computation-driven approach for guiding cancer genomics research and improving genomics-based clinical decisions in cancer treatment. The research activities described in the proposal will deliver a collection of effective and efficient computational tools to utilize heterogeneous genomic data combined with biomedical knowledge for clinical practices. The study of the ovarian cancer data will help reveal the crucial pathways driving chemoresistance, and provide useful prediction tools and drug targets for ovarian cancer treatment. This proposal will also integrate the latest research development in computational cancer genomics into new courses in several training programs to prepare students for their future professions to meet the need of workforce in the growing biomedical and health informatics industry in the upper midwest region. The education plan will also have a focus on recruiting students in minority and under-represented groups in computer science and information technology.To achieve the goals, the components of the research plan are 1) to formulate graph kernels and subgraph mining algorithms that can integrate various types of cancer genome aberrations to improve cancer outcome predictions and to discover cancer-causative genome aberration patterns; 2) to formulate semi-supervised matrix factorization methods with Laplacian constraints for predicting novel cancer phenotype and gene associations for identifying potential drug targets, utilizing known relations in phenotype, gene and their association networks; 3) to study the chemoresistance in ovarian cancer treatment to reveal the crucial pathways driving the resistance, and develop useful prediction tools and drug targets for ovarian cancer treatment; 4) to release the developed methods in both software packages and webtools for public use in academia. The two major components of the education plan are: 1) to offer a two-week course, titled Cure Cancer with Computers, in the summer academy of the BioSMART program for Minnesota high school students, and 2) to create a new course Computational Genomics in Biomedical Informatics to support two graduate programs for training students in biomedical/health informatics with knowledge in genomics and computer science.