Probability Model Based Clustering for Gene Activity States

As the quantity and diversity of transcriptomics data increase and the prices to obtain data continue to decline, numerous approaches which attempt to classify the gene activity into two states (active or inactive) appeared. These approaches suffer from many limitations, such as failing to overcome the limitation of data distribution. To address these limitations, we propose an improved Bayesian approach (KGMM) which combines k-means algorithm and Gaussian Mixture Model for the clustering of gene activity states. We run the approach we proposed and existing methods on both simulated data and real gene expression data. By comparing the performance of our approach to that of existing methods, we draw the conclusion that our approach yield more reasonable and accurate results than existing methods.