Bayesian model-based method for clustering gene expression time series with multiple replicates

In this study, we introduce a Bayesian model-based method for clustering transcriptomics time series data with multiple replicates. This technique is based on sampling Gaussian processes (GPs) within an infinite mixture model from a Dirichlet process (DP). Our method uses multiple GP models to accommodate for multiple differently behaving experimental replicates within each cluster. We call it multiple models Dirichlet process Gaussian process (MMDPGP). We compare our method with state-of-the-art model-based clustering approaches for handling gene expression time series with multiple replicates. We present a case study where all methods are applied for clustering RNA-Seq time series of Clostridium botulinum with three different experimental replicates. The results obtained from the gene enrichment analysis showed that the number of significantly enriched sets of genes is larger in the clusters produced by MMDPGP. To demonstrate the accuracy of our method we use it to cluster synthetically generated data sets. The clusters produced by our method on the synthetic data had a significantly higher purity score compared to the state-of-the-art approaches. By modelling each replicate with a separate GP, our method can use the natural variability between experimental replicates to learn more about the underlying biology. Author summaryIn our manuscript we introduce a method called multiple models Dirichlet process Gaussian process (MMDPGP), a novel Bayesian approach for clustering gene expression time series data. Our method stands out by accounting for the variability among multiple experimental replicates within each cluster, a feature that is often overlooked in existing model-based clustering approaches. This allows us to capture the natural variability between replicates as opposed to the crude method of simply averaging the replicates which discards interesting information in the data. By integrating multiple Gaussian process models within an infinite mixture model derived from a Dirichlet process, MMDPGP offers a more nuanced and accurate representation of the biological data. We benchmarked MMDPGP against state-of-the-art methods, by applying them for the purpose of clustering recently collected RNA-Seq time series of the bacterium Clostridium botulinum and performing a gene enrichment analysis on the generated clusters. Additionally, we test the accuracy of our method in comparison with other methods using synthetic data sets. The superior performance of our method in terms of finding significantly enriched gene sets and the clustering accuracy on synthetic data underscore its robustness and potential for broad applicability in computational biology. Our study addresses a critical gap in the analysis of transcriptomics time series data by explicitly modeling the natural variability across experimental replicates. This advancement not only enhances the accuracy of clustering results but also provides deeper insights into the underlying biological processes. By leveraging Bayesian methods and Gaussian processes, our approach offers a powerful tool that can be adapted and extended for various types of omics data, inspiring further methodological developments in the field. Competing interestsWe declare no competing interests related to this work. Code availability and implementationThe Python code for implementing our method is publicly available in Zenodo through the following DOI link: https://doi.org/10.5281/zenodo.11202145. DataThe RNA-Seq data used to validate our method in the paper are deposited in GEO at the following link: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE248529.