Shape-constrained, changepoint additive models for time series omics data with cpam

Abstract

Time series omics experiments are critical for the study of a wide range of biological processes such as cell differentiation and developmental programs or responses to pathogens and environmental cues. While statistical tools for differential analysis across static conditions have matured, comparable comprehensive methodology is lacking for time series data. Here, we introduce cpam, a novel time series method and user-friendly software that performs temporal differential analysis of omics data for case-only or case-control time series, incorporating quantification uncertainty. Powerful features of cpam include changepoint detection and shape-constrained temporal trend estimation to allocate omics data into similar clusters. Performance evaluation shows that cpam outperforms existing time series methods in terms of control of the false discovery rate versus power to detect temporal changes and accurate changepoint estimation. In particular, we find that trend-based models provide superior performance compared to pairwise comparisons, even with as few as six time points. The software provides an interactive interface with modern customizable visualisations, offering both graphical and statistical insight into complex molecular processes. Application to published transcriptome data illustrates RNA isoform-level modelling together with high resolution clustering during human embryogenesis and the identification of 910 novel genes differentially expressed in response to excess light in Arabidopsis. This method has the potential for application to a range of omics data, including single-cell analyses via pseudobulked data, providing the analytical power to reveal complex temporal patterns of gene expression in any biological system.