基于高效预测模型的原核精细调控元件理性设计

项目名称： 基于高效预测模型的原核精细调控元件理性设计

项目编号： No.31301017

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 生物科学

项目作者： 蒙海林

作者单位： 中国科学院上海生命科学研究院

项目金额： 25万元

中文摘要： 合成生命系统的设计需要预先定义和构建具有不同预期参数特征的精细调控元件，如启动子、核糖体结合位点(RBS)等，以适应复杂生物网络及途径设计对元件强度特性的多样化需求。尽管基于随机突变的元件文库构建方式已得到实际应用，但在大规模生物网络及系统设计中需要更理性的设计方式来提高效率和降低成本。基于计算及模型预测的元件序列设计因而将成为今后的发展趋势。而现有模型普遍采用基于线性回归分析的建模方法难以表征元件序列与强度(活性)之间所存在的异常复杂的非线性关系，从而导致预测准确率偏低，推广性差，无法真正实现调控元件理性设计的需求。本项目针对目前原核调控元件(启动子/RBS)强度预测及序列设计所存在的这些关键技术瓶颈，拟采用非线性映射能力很强的人工神经网络和支持向量机方法来构建高性能预测模型，并基于模型的预测，通过干湿实验相结合而实现原核精细调控元件的理性设计。

中文关键词： 定量设计；原核精细调控元件；人工神经网络；支持向量机；强度预测模型

英文摘要： The design of complex synthetic biosystems requires predefining and constructing a number of fine-tuning regulatory elements with various parameter characteristics, e.g., promoters and ribosome binding sites (RBSs) with a range of strengths. Although random mutation based library construction has practical application, more rational strategies are still required to improve efficiency and save costs during the design of large scale networks and systems. Therefore, developing methodologies based on model calculation and prediction for designing element sequence will become a trend in the future. However, existing models cannot achieve rational design of regulatory element sequence with desired strength, mostly due to the common use of linear regression analysis based modelling methods, which could not well reflect the extremely complex non-linear relationship between element sequences and their strengths, resulting in a low prediction accuracy and poor generality. In this project, to address these issues, artificial neural network and support vector machine based modelling methods with stronger non-linear mapping properties will be apply to construct predicting models with high performance, and finally succeed in rational design of fine-tuning regulatory elements through the combination of wet and dry experiments

英文关键词： Quantitative design；prokaryotic fine-tuning regulatory element；artificial neural network；support vector machine；strength prediction model