Deep learning of drug sensitivity and genetic dependency of pediatric cancer cells

儿科癌细胞药物敏感性和遗传依赖性的深度学习

• 批准号: 10657820
• 负责人: Yu-Chiao Chiu
• 金额: $ 24.9万
• 依托单位: UNIVERSITY OF PITTSBURGH AT PITTSBURGH
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-03-01 至 2025-06-30
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10657820
• 关键词: Adult; Antineoplastic Agents; Architecture; Area; Award; Bioinformatics; Biological Assay; Cancer cell line; Cause of Death; Cells; Characteristics; Chemicals; Child; Childhood; Clustered Regularly Interspaced Short Palindromic Repeats; Communities; Computer Models; Computing Methodologies; Data; Dependence; Development; Disease; Educational process of instructing; Environment; Future; Genetic; Genetic Predisposition to Disease; Genetic study; Genomics; Grant; In Vitro; Intelligence; Knowledge; Learning; Machine Learning; Malignant Childhood Neoplasm; Malignant Neoplasms; Mentors; Methods; Modeling; Modernization; Molecular Profiling; Mutation; Pattern; Pediatric Neoplasm; Performance; Pharmaceutical Preparations; Pharmacogenomics; Pharmacotherapy; Phase; Postdoctoral Fellow; Preclinical Testing; Publishing; Research; Research Personnel; Research Training; Resources; Sampling; Scheme; Seasons; Structure; Testing; Training; Training Activity; Translations; Writing; anticancer research; cancer cell; cancer genome; cancer genomics; career; clinically relevant; coping; cost; data resource; deep learning; deep learning model; design; drug discovery; drug sensitivity; drug testing; experimental study; gene network; genetic signature; genome-wide; genomic profiles; genomic signature; high dimensionality; high throughput analysis; high throughput technology; in vivo; innovation; insight; knockout gene; learning strategy; new therapeutic target; novel; novel therapeutics; patient derived xenograft model; pre-clinical; predicting response; predictive modeling; programs; resistance mechanism; response; skills; small molecule; transfer learning; tumor; tumor heterogeneity

Summary/Abstract The development of novel therapies for pediatric cancers, the second leading cause of death in children, is challenging due to the lack of comprehensive pharmacogenomics resources, unlike the well-established ones in adult cancers. However, breakthroughs in deep learning methods allow learning of intricate pharmacogenomics patterns with unprecedented performance. With a uniquely cross-disciplinary background, the candidate for this proposed K99/R00 has already, as a postdoctoral fellow, (i) developed and published several deep learning models that accurately predicted adult cancer cells’ drug sensitivity and genetic dependency using high- throughput genomics profiles, and (ii) demonstrated the feasibility of transferring the model to predict tumors by a ‘transfer learning’ design. The candidate will extend this research to study pediatric cancers and test the central hypothesis that deep learning extracts genomics signatures to predict the responses of pediatric cancer cells to chemical and genetic perturbations. The proposed study will develop novel deep learning models for predicting drug sensitivity and/or genetic dependency for (Aim 1) currently un-screened pediatric cancer cell lines by learning from screens of adult cells, and (Aim 2) pediatric tumors by learning from adult and/or pediatric cells. Prediction results will be validated by in vitro experiments and data collected from patient-derived xenografts. The proposed study is the first attempt to employ modern computational methods to advance pharmacogenomics studies of pediatric cancer, which would be difficult and costly to pursue via biological assays. Findings will shed light on the optimal drugs and novel therapeutic targets for pediatric malignancies, leading to an optimal and efficient design of preclinical tests. The candidate has a remarkable track record of bioinformatics studies of adult cancer genomics. The focus of this K99 training plan is to develop in-depth understanding of pediatric cancer and preclinical treatment models, and strengthen multifaceted components needed for a successful research career in cancer bioinformatics. The primary mentor, Dr. Peter Houghton, is a renowned leader in pediatric cancer research and preclinical drug testing programs. The candidate also has assembled an outstanding mentor team: Dr. Yidong Chen (co-mentor), a cancer genomics expert and pioneer in bioinformatics analysis of high-throughput technologies; Dr. Jinghui Zhang (collaborator), a computational biologist and leader in integrative genomics studies of major pediatric cancer genome consortiums; Dr. Yufei Huang (collaborator), an expert in state-of-the-art deep learning methods; and two highly knowledgeable consultants with relevant expertise. With this team’s guidance and structured training activities in an ideal training environment, the candidate will strengthen his skills in grant writing and lab management, teaching and mentoring, and broad connections. Overall, the K99/R00 award will be an indispensable support for a timely transition of the candidate to a successful career as a multifaceted, cross-disciplinary investigator in cancer bioinformatics.

总结/摘要 儿童癌症是儿童死亡的第二大原因， 由于缺乏全面的药物基因组学资源，这一研究具有挑战性， 成人癌症然而，深度学习方法的突破允许学习复杂的药物基因组学 以前所未有的性能模式。凭借独特的跨学科背景， K99/R 00作为博士后研究员，已经（i）开发并发表了几个深度学习 这些模型准确预测了成年癌细胞的药物敏感性和遗传依赖性， 通量基因组学谱，和（ii）证明了转移模型预测肿瘤的可行性， 一种“迁移学习”的设计。候选人将扩展这项研究，以研究儿童癌症和测试中央 假设深度学习提取基因组学特征来预测儿科癌细胞对 化学和基因干扰。这项研究将开发新的深度学习模型，用于预测 药物敏感性和/或遗传依赖性（目标1），目前未筛选的儿科癌细胞系， 从成人细胞的筛选中学习，和（目的2）通过从成人和/或儿科细胞学习儿科肿瘤。 预测结果将通过体外实验和从患者来源的异种移植物收集的数据进行验证。 这项研究是首次尝试采用现代计算方法来推进 儿科癌症的药物基因组学研究，这将是困难和昂贵的追求通过生物测定。 研究结果将揭示儿童恶性肿瘤的最佳药物和新的治疗靶点， 临床前试验的最佳和有效设计。这位候选人在生物信息学方面有着出色的记录 成人癌症基因组学的研究。本K99培训计划的重点是深入了解 儿科癌症和临床前治疗模式，并加强所需的多方面组成部分， 在癌症生物信息学方面的成功研究生涯。主要的导师，彼得霍顿博士，是一个著名的 儿科癌症研究和临床前药物测试项目的领导者。这位候选人还召集了一个 杰出的导师团队：癌症基因组学专家和生物信息学先驱陈怡东博士（联合导师） 高通量技术分析;计算生物学家和领导者Jinghui Zhang博士（合作者） 主要儿科癌症基因组联合体的整合基因组学研究;黄宇飞博士（合作者）， 一位最先进的深度学习方法专家;以及两位知识渊博的顾问， 专业知识在这个团队的指导下，在理想的培训环境中进行结构化的培训活动， 候选人将加强他在拨款写作和实验室管理，教学和指导，以及广泛的技能 连接.总的来说，K99/R 00奖将是候选人及时过渡的不可或缺的支持 作为一个多方面的，跨学科的癌症生物信息学研究者的成功职业生涯。