摘要： 针对肿瘤基表达谱样本少、维数高的特点，提出一种用于肿瘤分型的粒子群神经网络集成算法。根据相似性度量函数滤出分类无关基因，形成候选特征子集。采用基于灵敏度分析的BP神经网络模型作为基分类器，进一步剔除冗余基因。改进的粒子群优化算法全局搜索BP神经网络的权值和阈值。实验结果表明，该算法对肿瘤分型具有良好的识别率，且特征集合中仅包含54个特征基因。

关键词: 粒子群优化, 神经网络集成, 基因表达谱, 特征基因, 肿瘤分型

Abstract: Due to the characteristic of small sample numbers and high dimensionality in tumor gene expression profile, an ensemble algorithm of particle swarm neural network is proposed to classify tumor subtypes. The genes irrelevant to classification are eliminated by different correlation functions and candidate feature subsets are formed. BP neural network based on sensitivity analysis is used as base classifier to learn the subsets and redundant genes are further removed. The parameters and thresholds of classifiers are optimized by improved Particle Swarm Optimization(PSO) algorithm. Experimental results show that the proposed method can obtain better recognition rates in tumor classification and only 54 feature genes in the feature set.

Key words: Particle Swarm Optimization(PSO), neural network ensemble, gene expression profile, feature gene, tumor classification

中图分类号: 

• TP18