摘要
氨基酸含量是影响蛋白质耐热性的主要因素,为了提高以氨基酸含量为特征向量的蛋白质耐热性预测的精度和预测模型的性能,提出了一种基于机器学习蚁群算法(ACO)优化支持向量机(SVM)参数的蛋白质耐热性预测方法。建立了SVM参数优化模型,探讨了基于网格划分策略的连续蚁群算法,通过对SVM的惩罚因子和径向基核函数的全局搜索,筛选出最优参数,使SVM的蛋白质耐热性预测率最优。结果表明:采用未优化的SVM建立的预测模型的蛋白质耐热性总预测率相对较低,约为76.5%,采用遗传算法优化预测模型参数后的预测率约为86.6%,采用ACO优化预测模型参数后预测率达到87.8%。采用ACO优化的SVM模型参数的寻优速度快,预测结果准确。
The content of amino acid is the main factor to influence the protein heat resisting property. To enhance the prediction's accuracy and predict model's performance of the protein heat resisting property which utilizes the contents of amino acid aminophanol as its eigen vector. A predict method of the protein heat resisting property based on machine learning ACO algorithm which optimizes support SVM'S parameter is introduced. And a model of SVM'S parameter optimization is designed and the continual ACO algorithm based on the mesh generalization is discussed. By the way of global search to the penalty factor and radial basis fimction of SVM, the optimal parameter is screened out. Thus makes the protein heat resisting property rate of SVM optimum. Results indicate that the total prediction rate amounting to 76.5 % is relatively lower when utilizes the predict model build up by the non-optimized SVM. It is about 86.6 % when the genetic algorithm is adopted to optimize the predict model parameter. While it increased to 87.8 % when they predict model parameter is optimized with ACO. The optimal speed of the SVM model optimized with ACO is fast, and the prediction is accuracy.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2014年第2期166-170,共5页
Computers and Applied Chemistry
基金
国家自然科学基金资助项目(21001053)