基于多核LSSVM的谷物蛋白质二级结构预测与优化

Research on Prediction and Optimization of Cereal Protein Secondary Structure Based on Multi-Kernel LSSVM

摘要:

蛋白质的二级结构对其空间结构和功能有着极其重要的影响，利用机器学习方法进行谷物蛋白质二级结构预测是生物和食品领域的重要研究内容。作者在现有蛋白质数据库中选取玉米、小麦、大豆的谷物蛋白质，使用多特征融合方式对蛋白质序列进行特征提取，提出将多核学习与最小二乘支持向量机（LSSVM）相结合，以多个核函数的线性加权组合代替传统单一核函数，利用核权重调整融合效果，构建多核LSSVM模型预测谷物蛋白质二级结构。使用粒子群优化算法（PSO）对模型超参数进行优化，寻找最佳超参数组合提升模型预测性能。研究结果表明，多核LSSVM模型能够改善单一核函数高维映射的局限性，融合各核函数优势，通过PSO算法获取最佳超参数组合。该模型结合多特征提取方式显著提高了谷物蛋白质二级结构预测的Q₃准确率。

Abstract:

The secondary structure of protein has an extremely important impact on their spatial structure and function. Using machine learning methods to predict the secondary structure of grain proteins is one of the important topics in the fields of biology, food, etc.In this experiment, three grain proteins of corn, wheat, and soybeans were selected from the existing protein data bank,using multi-feature fusion method to extract features of protein sequence, and propose to combine multi-core learning with least square support vector machine classifier, the linear weighted combination of multiple kernel functions is used to replace the traditional single kernel function, and the kernel weight is used to adjust the fusion effect to construct a multi-core LSSVM prediction model to predict the secondary structure of grain proteins.The particle swarm optimization algorithm is used to optimize the model hyper parameters and find the best hyper parameter combination to improve the prediction performance of the model. The experimental results prove that the multi-kernel LSSVM algorithm proposed in this paper can make up for the limitations of single kernel function high-dimensional mapping, integrate the advantages of each kernel function, and find the best hyperparameter combination through the PSO algorithm, combine multiple feature extraction methods to significantly improve the Q₃ accuracy of grain protein secondary structure prediction.