宫颈癌HPV分型检测的关键算法研究与应用

Human papillomavirus (HPV) genotyping and mutation analysis models have important significance for early detection and treatment of cervical cancer. The differences between biological sequences are the main basis of HPV genotyping model, how to efficiently extract local and global information of sequences , selection and integration of core information are the bottlenecks of building HPV genotyping model. By integrating multiple sources of information which are obtained from sequence analysis, this project aims to build an efficient model of cervical cancer HPV genotyping. Based on the nucleotide sequence of distribution, the local information of sequence can be obtained by defining the position of the distribution function and calculating the degree of overlap subsequence. Taken the sequence fragments as a Bernoulli experiment, the overall information extraction algorithm is designed based on the binomial distribution. Then, by improving the LZ complexity and designing cumulative contribution function, the HPV core information can be selected from multiple information sources which contain subsequence distribution, location information and structural information. According to the characteristics of HPV multi-source information, the fuzzy multilayer perceptron network machine learning algorithms can be used to design HPV subtype prediction model. HPV according to the characteristics of multi-source information, using fuzzy Multilayer Perceptron network machine learning algorithms, design combines the core message of HPV genotyping prediction algorithm, and validate the model. This project will not only help to guide early detection and treatment of cervical cancer, but also open up new avenues of molecular diagnostics and personalized treatment and prevention of other cancers.

人乳头瘤病毒(HPV)分型对宫颈癌的早期发现及指导治疗具有重要意义。生物序列之间的差异性是HPV分型的主要依据，如何有效地提取序列局部和整体信息，挑选和融合核心信息是构建高效HPV分型模型的瓶颈问题。本项目拟从HPV序列分析入手，融合多源信息构建高效的宫颈癌HPV分型模型:基于序列碱基分布特点，定义其位置分布函数，计算序列片段的重叠与保守程度，获取HPV序列的局部信息；基于位置分布函数，设计伯努利实验，提出基于二项分布的信息提取算法，获取序列的整体信息；改进LZ复杂度，设计累计贡献函数，从包含子序列分布、位置分布及结构信息的多源信息中挑选HPV核心信息；根据HPV多源信息的特点，利用模糊多层感知器网络等机器学习算法，设计融合核心信息的HPV分型预测算法，并对该模型加以验证。本项目的深入研究不仅有助于宫颈癌早期发现及指导治疗，也为其他肿瘤的分子诊断与防治以及个性化治疗开辟全新的途径。

人乳头瘤病毒(HPV)分型对宫颈癌的早期发现及指导治疗具有重要意义。生物序列之间的差异性是HPV分型的主要依据，如何有效地提取序列局部和整体信息，挑选和融合核心信息是构建高效HPV分型模型的瓶颈问题。本项目拟从HPV序列分析入手，融合多源信息构建高效的宫颈癌HPV分型模型:基于序列碱基分布特点，定义其位置分布函数，计算序列片段的重叠与保守程度，获取HPV序列的局部信息；基于位置分布函数，设计伯努利实验，提出基于二项分布的信息提取算法，获取序列的整体信息；改进LZ复杂度，设计累计贡献函数，从包含子序列分布、位置分布及结构信息的多源信息中挑选HPV核心信息；根据HPV多源信息的特点，利用模糊多层感知器网络等机器学习算法，设计融合核心信息的HPV分型预测算法，并对该模型加以验证。本项目的深入研究不仅有助于宫颈癌早期发现及指导治疗，也为其他肿瘤的分子诊断与防治以及个性化治疗开辟全新的途径。课题组已在Swarm and Evolutionary Computation、Neurocomputing和Frontiers in Genetics等 SCI 杂志上发表学术论文 25 篇，获得国家自然科学基金资助项目3项，省自然科学基金资助1项，培养研究生12名，承办国际学术会议1次，国内会议2次。