基于PaDEL分子描述符的气味机器学习识别研究
Odor molecule recognition via PaDEL molecular descriptors and machine learning
苏洋洋 1夏仁杰 1王云松 1许振军2
作者信息
- 1. 江苏理工学院电气信息工程学院,江苏常州 213001
- 2. 浙江古越龙山电子科技发展有限公司,浙江绍兴 312000
- 折叠
摘要
气味识别有助于更好地理解嗅觉处理机制和生物进化中的交流方式,也是食品、饮料和香水等商业领域的关键技术.在利用PaDEL对气味分子进行数值化表征的基础上,采用决策树、随机森林、极端随机树、支持向量机、全连接神经网络和循环神经网络等机器学习算法,对PaDEL分子描述符进行了特征学习,以实现对气味的多标签分类.结果显示,FCNN和DT模型的预测性能较强,BiGRU模型的预测性能最弱.采用主成分分析算法对PaDEL数据进行了降维处理,基于此数据的不同机器学习模型气味分类性能比较中,BiGRU模型的性能最优.
Abstract
Odor recognition contributes to a better understanding of olfactory processing mechanisms and com-munication modes in biological evolution,and is a key technology in various commercial sectors such as food,beverages,and perfumes.Based on the numerical characterization of odor molecules using PaDEL,feature learn-ing of PaDEL molecular descriptors is carried out by using machine learning algorithms such as decision tree,random forest,extreme random tree,support vector machine,fully connected neural network and recurrent neu-ral network to realize multi-label classification of odor.The results show that the predictive performance of FCNN and DT models is relatively good,while the predictive performance of BiGRU model is worst.The principal com-ponent analysis was utilized to reduce the dimensionality of the PaDEL data,and in the comparison of odor classi-fication performances of different machine learning models based on this data,BiGRU model is the best.
关键词
PaDEL分子描述符/机器学习/气味识别Key words
PaDEL molecular descriptors/machine learning/odor recognition引用本文复制引用
基金项目
常州市基础研究计划(应用基础研究)项目(CJ20200045)
江苏省青年自然科学基金项目(BK20191032)
出版年
2024
NETL
NSTL
万方数据