Modeling cancer evolution for prediction with neural networks: methods and applications

项目总结/摘要 对肿瘤演变的研究可以揭示推动肿瘤发展的事件和相互作用 通过替代途径，揭示治疗弱点的差异，并改善临床决策 制作。然而，研究肿瘤进化是具有挑战性的，由于难以解释嘈杂的基因组序列， 数据以及缺乏主要遗传事件的时间顺序。因此，迫切需要 发展计算方法，可以促进癌症数据的有效调查， 一个进化的、暂时的视角。这个项目的长期目标是开发计算工具 将先进的机器学习与分子进化技术相结合， 来研究肿瘤的演变总体目标是建立一个深度学习框架， 肿瘤的发展，将用于区分早期和晚期的遗传事件，与 肿瘤特征、存活率和治疗脆弱性。拟议研究的基本原理是 通过整合机器学习和分子进化来研究肿瘤进化 这些技术可以增强否则困难的临床分类任务的性能。具体 本项目的目的是（1）描述肿瘤中驱动突变和非整倍体之间的相互作用， 演变和确定临床结果和治疗脆弱性的决定因素（2）介绍 通过时间和功能排序表示快照基因组数据的计算方法 （3）开发一种循环神经网络方法来学习肿瘤中的不同动力学 从有序的基因组数据进化，并预测表型特征和临床结果。的 拟议的研究是创新的，因为它将联合收割机在机器学习方面的最新进展与 进化技术到一个单一的框架，建立新的计算工具，将促进 对癌症发展的全面调查拟议的框架意义重大，因为它 将使时间建模与机器学习应用于癌症数据，用于预测临床 功能.为了实现拟议的目标，候选人Noam Auslander博士需要接受额外的培训 并在进化研究、比较基因组学和数学建模方面提供指导。期间 K99阶段，Auslander博士将在国家中心作为博士后研究员进行这项研究， 生物技术信息.她将由该领域公认的专家尤金库宁博士指导 分子进化和计算生物学，以及来自 库宁实验室加上她之前在机器学习和癌症数据科学方面的培训， 申请NIH独立之路奖（K99/R 00）描述了职业发展计划 这将使Auslander博士能够实现她的职业目标，成为一名独立的调查员， 癌症进化计算研究的领导者。

PROJECT SUMMARY/ABSTRACT The study of tumor evolution can uncover events and interactions that drive tumor development through alternative routes, reveal differences in therapeutic vulnerabilities and improve clinical decision making. Yet, studying tumor evolution is challenging, hindered by the difficulty to interpret noisy genomic data and the lack of temporal ordering of major genetic events. There is therefore a critical need for the development of computational approaches that can facilitate efficient investigation of cancer data under an evolutionary, temporal perspective. The long-term goal of this project is to develop computational tools combining advanced machine learning with molecular evolution techniques and provide novel strategies to investigate tumor evolution. The overall objective is to establish a deep-learning framework to study tumor development that will be used to distinguish early and late genetic events that are associated with tumor characteristics, survival and therapeutic vulnerabilities. The rationale of the proposed research is that the study of tumor evolution through integration of machine learning with molecular evolution techniques could enhance the performance of otherwise difficult clinical classification tasks. The specific aims of this project are to (1) Characterize the interplay between driver mutations and aneuploidy in tumor evolution and identify determinants of clinical outcome and therapeutic vulnerabilities (2) Introduce computational approaches to represent snapshot genomic data through temporal and functional ordering of genetic events (3) Develop a recurrent neural network approach to learn different dynamics in tumor evolution from ordered genomic data, and predict phenotypic features and clinical outcome. The proposed research is innovative because it will combine recent advances in machine learning with evolutionary techniques into a single framework, establishing novel computational tools that will facilitate a comprehensive investigation of cancer development. The proposed framework is significant because it will enable application of temporal modeling with machine learning to cancer data, for prediction of clinical features. To achieve the proposed goals the candidate, Dr. Noam Auslander, requires additional training and mentoring in evolutionary research, comparative genomics and mathematical modeling. During the K99 phase, Dr. Auslander will conduct this research as a postdoctoral fellow at the National Center for Biotechnology Information. She will be mentored by Dr. Eugene Koonin, a recognized expert in the fields of molecular evolution and computational biology, and additional mentoring from senior members from the Koonin lab. Together with her previous training in machine learning and cancer data science, this application for the NIH Pathway to Independence Award (K99/R00) describes a career development plan that will allow Dr. Auslander to achieve her career goals and become an independent investigator and leader in computational research of cancer evolution.