肿瘤亚型识别研究中智能算法的应用

目的:为解决肿瘤亚型识别过程中易出现的维数灾难和过拟合问题,提出了一种改进的粒子群BP神经网络集成算法。方法:算法采用欧式距离和互信息来初步过滤冗余基因,之后用Relief算法进一步处理,得到候选特征基因集合。采用BP神经网络作为基分类器,将特征基因提取与分类器训练相结合,改进的粒子群对其权值和阈值进行全局搜索优化。结果:当隐含层神经元个数为5时,候选特征基因个数为110时,QPSO/BP算法全局优化和搜索,此时的分类准确率最高。结论:该算法不但提高了肿瘤分型识别的准确率,而且降低了学习的复杂度。

Application of An Intelligent Algorithm in Tumor Subtype Recognition

ABSTRACT Objective: In order to solve the dimension disaster and over-fitting problems in the process of tumor subtype recogni- tion, a particle swarm optimization (PSO) BP neural network ensemble algorithm was proposed. Methods: The Euclidean distance and mutual information was used to preliminarily filter redundant genes, and then Relief algorithm was adopted to further process the candi- date feature genes set. The BP neural network was used as the base classifier, which combines feature genes extraction with classifier training. Results: When the number of hidden layer neurons is 5 and the number of candidate feature genes is 110, the QPSO/BP algo- rithm can optimize and search globally. Conclusion: The algorithm not only improves the accuracy of tumor classification and recogni- tion, but also reduces the complexity of learning.