kmmDE: A Non-Parametric Method for Differential Expression Analysis of Time-Course RNA-Seq Data Using Maximum Mean Discrepancy

Accurately identifying differentially expressed genes from time-course RNA-Seq data is essential for understanding the dynamics of transcriptome changes during biological processes. Although various methods for differential expression analysis of time-course RNA-seq data were developed, they are limited by their reliance on specific distribution assumptions, and their performance is restricted in datasets without replicates or in short time-series datasets. To address these challenges, we propose a non-parametric method called kmmDE for differential expression analysis based on maximum mean discrepancy. kmmDE utilizes kernel functions to map data samples into a high-dimensional feature space, where it assesses the similarity between two distributions by comparing their mean embeddings. kmmDE can adapt to any form of data distribution without the need to presuppose distribution shapes. The evaluation of kmmDE on simulated datasets and comparison with six popular methods showed improved results for experiments with or without biological replicates. Moreover, our real data analysis of grape cold-resistant time-course RNA-Seq data demonstrates that kmmDE identifies biologically relevant genes which are well justified by Gene Set Enrichment Analysis. This study provides a useful tool for differential expression analysis of time-course RNA-Seq data, particularly for datasets lacking replicates or with short time series. The source code and data are available at https://github.com/zhanglab-wbgcas/kmmDE.