[论文解读] Verdict Accuracy of Quick Reduct Algorithm using Clustering and Classification Techniques for Gene Expression Data

T. Chandrasekhar, K. Thangavel+1 more|arXiv (Cornell University)|Jun 6, 2013

Gene expression and cancer classification参考文献 12被引用 9

一句话总结

本文提出了一种基于粗糙集理论的快速约简算法（Quick Reduct）的混合特征选择与分类方法，用于基因表达数据，随后结合K-均值和模糊C-均值聚类，以及反向传播网络（BPN）分类。该方法识别出一个最小的有信息量基因集，且BPN在分类准确率上优于仅使用聚类的方法，展示了在诊断预测任务中更高的判断准确率。

ABSTRACT

In most gene expression data, the number of training samples is very small compared to the large number of genes involved in the experiments. However, among the large amount of genes, only a small fraction is effective for performing a certain task. Furthermore, a small subset of genes is desirable in developing gene expression based diagnostic tools for delivering reliable and understandable results. With the gene selection results, the cost of biological experiment and decision can be greatly reduced by analyzing only the marker genes. An important application of gene expression data in functional genomics is to classify samples according to their gene expression profiles. Feature selection (FS) is a process which attempts to select more informative features. It is one of the important steps in knowledge discovery. Conventional supervised FS methods evaluate various feature subsets using an evaluation function or metric to select only those features which are related to the decision classes of the data under consideration. This paper studies a feature selection method based on rough set theory. Further K-Means, Fuzzy C-Means (FCM) algorithm have implemented for the reduced feature set without considering class labels. Then the obtained results are compared with the original class labels. Back Propagation Network (BPN) has also been used for classification. Then the performance of K-Means, FCM, and BPN are analyzed through the confusion matrix. It is found that the BPN is performing well comparatively.

研究动机与目标

• 通过选择最小且有信息量的基因子集，解决高维基因表达数据样本较少的挑战。
• 通过特征选择识别标志基因，提高诊断准确率并降低实验成本。
• 评估在降维基因集上聚类与分类技术的性能，以实现可靠的样本分类。
• 比较K-均值、模糊C-均值和反向传播网络在所选基因特征上的分类准确率。

提出的方法

• 应用快速约简算法对基因表达数据进行处理，基于粗糙集理论提取最小且相关的特征子集。
• 在降维后的特征子集上应用K-均值和模糊C-均值（FCM）聚类，不使用类别标签以识别模式。
• 将聚类结果与原始类别标签进行比较，评估降维表示的一致性与准确性。
• 在降维后的特征子集上训练反向传播网络（BPN）以进行分类与性能评估。
• 使用混淆矩阵评估各类方法的分类准确率，进行比较。
• 通过将各方法的预测类别标签与真实标签对比，确定最终判断准确率。

实验结果

研究问题

• RQ1 快速约简算法在保留诊断相关性的同时，对基因表达数据的降维效果如何？
• RQ2 在无类别标签信息的情况下，K-均值和FCM等聚类技术能否在降维基因集中识别出有意义的模式？
• RQ3 BPN在降维基因子集上的分类准确率与无监督聚类方法相比如何？
• RQ4 基于粗糙集理论的特征选择在多大程度上提升了基因表达分类的整体诊断准确率？

主要发现

• 反向传播网络（BPN）在所评估方法中取得了最高的分类准确率，优于K-均值和模糊C-均值。
• 快速约简算法成功识别出一个最小且有信息量的基因子集，在降低维度的同时保持了诊断相关性。
• 使用K-均值和FCM的聚类结果与原始类别标签表现出中等程度的一致性，表明在降维空间中部分结构得以恢复。
• 混淆矩阵分析证实，BPN在降维基因集上提供了最一致且准确的预测。
• 将基于粗糙集的特征选择与BPN分类相结合，显著提升了基因表达诊断中的判断准确率。