Machine learning models highlight environmental and genetic factors associated with the Arabidopsis circadian clock.

Abstract

The circadian clock of plants contributes to their survival and fitness. However, understanding clock function at the transcriptome level and its response to the environment requires assaying across high resolution time-course experiments. Generating these datasets is labour-intensive, costly and, in most cases, performed under tightly controlled laboratory conditions. To overcome these barriers, we have developed ChronoGauge: an ensemble model that can reliably estimate the endogenous circadian time of Arabidopsis plants using the expression of a handful of time-indicating genes within a single time-pointed transcriptomic sample. ChronoGauge can predict a plant's circadian time with high accuracy across unseen Arabidopsis bulk RNA-seq and microarray samples, and can be further applied to make non-random predictions across samples in non-model species, including field samples. Finally, we demonstrate how ChronoGauge can be applied to generate hypotheses regarding the response of the circadian transcriptome to specific genotypes or environmental conditions.