基于多检测理论融合的基因组结构变异综合检测方法

项目名称： 基于多检测理论融合的基因组结构变异综合检测方法

项目编号： No.61472026

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 高敬阳

作者单位： 北京化工大学

项目金额： 80万元

中文摘要： 精准检测基因组结构变异，对变异形成机制的研究、揭示人类基因变异与复杂疾病之间的关系非常重要。现有基于测序的结构变异检测方法存在的问题是每种检测方法均有不同的各自侧重点，均依赖于结构变异的类型或者序列片段的特性，而且对各种检测方法的评价没有公认的金标准。本项目以现有的结构变异检测方法深度分析为出发点，重点研究融合多种检测理论的结构变异综合表征和适应大样本学习的高泛化性能复杂网络模型，突破特征指标体系、特征参数敏感性规律、学习训练样本集、基于争议度的AdaBoost权值调整策略和神经网络逆向权值调整策略等关键技术，建立一套先进的基因组结构变异综合检测的理论与技术体系。本项目的研究结果将有利于对基因结构变异分类产生新的认识，在基因组水平进行大规模的结构变异发现、识别结构变异类型、建立人类的结构变异图谱资源方面具有十分积极的意义。

中文关键词： 基因组结构变异；多特征融合；神经网络集成模型；检测方法

英文摘要： Accurate detection of genomic structural variation is very important to study mutation mechanistic origins and reveal the relationship between human genetic variation and complex diseases. Existing sequencing-based structural variation discovery methods focus differently and they mainly rely on the types of variants and the features of reads. Furthermore, it is still lack of a 'gold standard' for evaluation of disparate methods. Starting with deep analysis of the state-of-the-art structural variation detection approaches, by making breakthrough in setting up characteristics index system, finding characteristics parameters sensitivity laws, improving ERstd-AdaBoost and Inverse Boosting algorithms, this project finally constructs a theoretical and technological framework for structural variation detection, which consists of an integrated structural variation characterization model based on varieties of discovery theories and a complex network model of high generalization performance adapting to large sample study. This study will propose a new way to understand structural variation classification and have great significance in large-scale genomic structural variation discovery and classification as well as establishment of human structural variation map.

英文关键词： genomic structural variation;multi-feature fusion;neural network ensemble model;detection method