Identification and correction of time-series transcriptomic anomalies

Abstract

Transcriptomic analyses performed in time series have uncovered many important insights into dynamic biological processes such as circadian rhythms, cellular developmental cycles, and the cell cycle. Some of these studies have revealed transcriptomic artifacts (STRIPEs), characterized by substantial changes in transcript levels across the transcriptome between a time point and its temporal neighbors. These changes are unlikely to reflect underlying biology as the magnitude of the change is too large to occur within the time interval and because every gene in the time point exhibits a substantial change. Furthermore, STRIPEs occur across species exhibiting different biology, do not occur in the same phase across replicate time-series experiments, and can vary between technical replicates of a single time point. Here, we demonstrate STRIPEs in five time-series transcriptomic datasets across different species, biological processes, and timescales. We describe a computational method to detect STRIPEs in time series using the Kolmogorov–Smirnov statistical test, allowing for unbiased, user-friendly detection of STRIPEs. Finally, we present three methods for STRIPE correction and demonstrate their efficacy. Using periodicity analysis to identify periodic genes, we find nearly 600 genes changed in periodicity labeling following successful STRIPE correction, indicating the large impact of STRIPE removal on downstream analysis.