癌症相关磷酸化信号转导网络攻击变异的计算识别和分析研究

Network attacking mutations (NAMs) affecting phosphorylation signaling are prevalent in many common malignant cancer types such as breast cancer, and are very crucial in illustrating the pathogenesis of cancers and developing efficient approaches for diagnosis and treatment. Given the present condition that current computational approaches are less efficient and scarce, this project aims to investigate novel methodologies and ideas in the identification and analysis of NAMs attacking phosphorylation sites and protein kinases, which are now attracting a lot of attention in cancer research. We plan to study multi-modal deep learning methods for sequence motifs of various PTM types such as phosphorylation and ubiquitination, and further develop novel identification approaches for NAMs that attack phosphorylation sites directly or mediated by PTM crosstalk. To solve the problem of identification of CNA-related NAMs that attack kinases, we propose a novel idea by developing joint matrix factorization algorithms for both cancer genome and phosphoproteome data. Based on the aforementioned studies, we adopt systematic analysis of multiple types of NAMs as a breakthrough and try to illustrate the characteristic and rules of attacks of NAMs to phosphorylation signaling, and the associations between NAMs and clinical outcomes as well as the sensitivities of anti-cancer drugs. With great theoretical significance and application value, this project will provide crucial computational and analytic approaches for studies of NAMs in cancer, and will be of great help for studies of the molecular mechanisms of NAMs and other NAMs-related precision medicine researches.

磷酸化信号转导的网络攻击变异（NAMs）在乳腺癌等高发恶性癌症中广泛存在，并对揭示癌症致病机制及发展有效诊疗手段意义重大。本项目以备受关注的磷酸化位点NAMs和激酶NAMs为对象，针对现有计算手段低效、匮乏的现状，积极探索NAMs识别和分析的新方法与新思路。拟深入开展磷酸化、泛素化等多种类型PTM序列模体的多模态深度学习方法研究，在此基础上发展面向直接攻击和crosstalk介导攻击的新型磷酸化位点NAMs识别方法。针对CNA变异相关激酶NAMs的识别难题，率先提出联合癌症基因组和磷酸化蛋白质组数据的矩阵分解算法新思路。在上述研究基础上，以多种类型NAMs的系统分析为突破点，深入揭示NAMs攻击磷酸化信号转导的特点规律以及与临床生存、药物敏感性的联系。本项目将为癌症NAMs研究提供准确有效的计算分析手段，并有助于NAMs攻击的分子机制及相关精准医学的研究，具有重要的理论意义和应用价值。

针对磷酸化信号转导的网络攻击变异（NAMs）与多种恶性癌症的侵袭与转移、患者生存等临床表型密切相关，对揭示癌症病因机制和发展精准诊疗手段均具有重大价值。本项目从智能化信息处理的角度出发，对NAMs计算识别与临床应用中的一些关键问题进行了深入研究。针对多类型PTM模体学习与位点精确检测的难题，创新性地发展了多尺度密集连接网络及其衍生的神经网络结构设计，用于实现复杂序列模体的高效自主学习。同时，从层级化预测导致训练数据稀缺等实际问题出发，提出了切实有效的新型迁移学习策略，确保神经网络的有效训练与泛化能力提升。在上述技术手段的有力支撑下，建立了多种具有国际先进水平的PTM位点检测算法及新型NAMs识别手段。此外，提出了一种基于子空间学习的驱动基因变异检测方法，解决了TCGA癌症基因组变异中大量随机突变严重干扰NAMs识别的问题。在探索NAMs攻击信号转导通路特点的过程中，巧妙地引入了分子功能网络信息与深度学习集成方案，在提升NAMs相关位点识别精度的同时，促进对磷酸化信号转导中NAMs攻击机制的深入理解。在上述研究的基础上，积极开展了乳腺癌、脑胶质瘤、肺癌等多类恶性肿瘤中的NAMs关联分析，同时建立了NAMs攻击与癌症多组学、病理图像等异构多源临床信息的高效融合方法，在癌症患者的个体化生存预测中获得了成功应用。上述研究成果为NAMs计算与分析技术提供扎实的理论基础，对NAMs相关的分子机制研究和癌症精准医学应用提供了重要的技术支撑。