Fuzzy C means Based Evaluation Algorithms For Cancer Gene Expression Data Clustering

Abstract

The influx of data in bioinformatics is primarily in the form of DNA, RNA, and protein sequences. This condition places a significant burden on scientists and computers. Some genomics studies depend on clustering techniques to group similarly expressed genes into one cluster. Clustering is a type of unsupervised learning that can be used to divide unknown cluster data into clusters. The k-means and fuzzy c-means (FCM) algorithms are examples of algorithms that can be used for clustering. Consequently, clustering is a common approach that divides an input space into several homogeneous zones; it can be achieved using a variety of algorithms. This study used three models to cluster a brain tumor dataset. The first model uses FCM, which is used to cluster genes. FCM allows an object to belong to two or more clusters with a membership grade between zero and one and the sum of belonging to all clusters of each gene is equal to one. This paradigm is useful when dealing with microarray data. The total time required to implement the first model is 22.2589 s. The second model combines FCM and particle swarm optimization (PSO) to obtain better results. The hybrid algorithm, i.e., FCM–PSO, uses the DB index as objective function. The experimental results show that the proposed hybrid FCM–PSO method is effective. The total time of implementation of this model is 89.6087 s. The third model combines FCM with a genetic algorithm (GA) to obtain better results. This hybrid algorithm also uses the DB index as objective function. The experimental results show that the proposed hybrid FCM–GA method is effective. Its total time of implementation is 50.8021 s. In addition, this study uses cluster validity indexes to determine the best partitioning for the underlying data. Internal validity indexes include the Jaccard, Davies Bouldin, Dunn, Xie–Beni, and silhouette. Meanwhile, external validity indexes include Minkowski, adjusted Rand, and percentage of correctly categorized pairings. Experiments conducted on brain tumor gene expression data demonstrate that the techniques used in this study outperform traditional models in terms of stability and biological significance.