Bayesian approach for analysis of time-to-event data in plant biology

Abstract

Background: Plants, like all living organisms, metamorphose their bodies during their lifetime. All the developmental and growth events in a plant's life are connected to specific points in time, be it seed germination, seedling emergence, the appearance of the first leaf, heading, flowering, fruit ripening, wilting, or death. The onset of automated phenotyping methods has brought an explosion of such time-to-event data. Unfortunately, it has not been matched by an explosion of adequate data analysis methods. Results and discussion:In this paper, we introduce the Bayesian approach towards time-to-event data in plant biology. As a model example, we use seedling emergence data of maize under control and stress conditions but the Bayesian approach is suitable for any time-to-event data (see the examples above). In the proposed framework, we are able to answer key questions regarding plant emergence such as these: (1) Do seedlings treated with compound A emerge earlier than the control seedlings? (2) What is the probability of compound A increasing seedling emergence by at least 5 percent? Conclusion:Proper data analysis is a fundamental task of general interest in life sciences. Here, we present a novel method for the analysis of time-to-event data which is applicable to many plant developmental parameters measured in field or in laboratory conditions. In contrast to recent and classical approaches, our Bayesian computational method properly handles uncertainty in time-to-event data and it is capable to reliably answer questions that are difficult to address by classical methods.

Deep learning-based detection of seedling development

“…Lidar cameras, accessible at low-cost [48], could be used to access to night events. Also, Bayesian approaches [6], such as Gaussian processes, could be used to estimate the time for the possibly missing information.…”

“…At the seedling level where plants have simple architectures, such time-lapse imaging can be done from top view to provide an efficient solution for seedling vigor assessments and monitoring of seedling growth. While some statistical tools transferred from developmental biology exists to perform time-to-event analysis [6], a current bottleneck [7] lay in the automation of the image analysis. A recent revolution occurred in the field of automated image analysis with deep neural networks [8], which have shown their universal capability to address almost any image processing challenges with high accuracy.…”

Comparing factors affecting dental‐implant loss between age groups: A retrospective cohort study

“…In a recent study, the authors demonstrated better results were obtained using the Bayesian statistical method relative to classical statistical analysis 33 . It is proposed that, accounting for the posterior distribution, this method properly deals with uncertainty, offering more realistic results 34 . Therefore, further investigation of the present data using different statistical models may reveal more insightful results.…”

Mixed Models as a Tool for Comparing Groups of Time Series in Plant Sciences

“…[ 13 , 14 ]. As a high-throughput method, phenotyping produces a quantity of longitudinal or time-to-event data which is challenging to process by an adequate statistical approach [ 15 , 16 , 17 ]. In our study, we show that not considering the dependency that exists among individual time points in time series might lead to wrong assumptions, which might further cause a false hypothesis to be made.…”