Machine learning for mechanistic modeling of plant metabolic pathways.

Abstract

Advances in computational power and structured representation of biochemical knowledge have fostered the development and use of models of plant metabolic pathways of increasing complexity. These advances have pointed out knowledge gaps about parameters inherent to these models and the need for large-scale data sets to conduct effective model calibrations. This bottleneck hampers wider application of plant metabolic models in plant biotechnology and crop breeding. Here, we argue that the recent advances in artificial intelligence offer the means for efficient model parameterization while strategically guiding the use of resources for data generation to increase precision of parameter values. We focus on surrogate models that have been applied with both stoichiometric and kinetic metabolic models as well as recent approaches for machine learning approaches for model parameterization using different data inputs. The review does not only showcase applications with plant metabolic pathways, but also points at recent examples from non-plant and chemical systems that can be readily adopted in the study of plant metabolism. Finally, we point at limitations and future directions that may deepen the synergies between machine/deep learning and mechanistic metabolic models to enable predictions of and mitigating actions for plant responses adapted to future climate scenarios.