Use of transcriptomic data to inform biophysical models via Bayesian networks

Process-based models of plant productivity provide a means of yield prediction that can inform best practices in agriculture and land management. However, current biophysical models fail in capturing both genotypic and phenotypic variation under changing environmental conditions. Physiological traits affecting final yield and ecosystem productivity are the result of gene expression, protein translation, and metabolite formation which are controlled by specific alleles acting alone and in response to time and environmental cues. Biophysical interactions take place simultaneously across several scales of time and space, giving rise to high levels of complexity and limiting the predictive capacity of existing analytical approaches. While statistical methods can partially quantify genotype-by-environment connections, biophysical process models miss exploring genotypic diversity and they cannot currently be implemented with -omics data, i.e. the entire collection of entities such as gene transcripts, metabolites, and proteins. Here we present a novel framework that utilizes Bayesian networks to provide probabilities that link gene expression levels to trait occurrence. We propose the use of these gained probabilities to inform parameters of existing biophysical models. Merging transcriptomic information into process-based models allows for greater integration of empirical data and processes across different scales in complex systems, while increasing the likelihood of growth predictions under unknown environmental conditions.