AI fast-track for development of water saving plants
Understanding how plants control their rate of gas exchange and water loss is a fundamental part of plant biology. However, studying the tiny doughnut-shaped pores on leaves鈥攌nown as stomata鈥攗sing microscopy, can be time-consuming.
Story: Kurtis Eichler / AIML
In two papers published last month in and , researchers from the 最新糖心Vlog of Adelaide鈥檚 Waite Research Institute worked with the 最新糖心Vlogn Institute for Machine Learning to develop AI-powered software tools which can make analysing the dynamics of stomata less labour-intensive.
In both studies, researchers used a type of AI called deep computer vision to accurately determine the number of open and closed stomatal pores and their density on the leaf; as well as take precise measurements of a stomatal pore鈥檚 area, length, and aperture from microscope images.
Published in New Phytologist, software tool was developed by AIML machine learning PhD candidate James Bockman and Dr Na Sai during her plant physiology PhD.
Known to science since 1889, stomatal pores are vital to ecosystems as they act as gateways allowing carbon dioxide to enter a plant and oxygen and water to exit.
Manually measuring the aperture of plant stomata allows researchers to get a clearer picture of how they function and respond to their environment. It is a common task for biologists studying plant signalling pathways and stress perception.
While some computer programs have attempted to semi-automate this process in the past, no such technology has allowed researchers to get results seconds after microscopy images are taken.
StomaAI has been initially trained to work on the crop plant barley and Arabidopsis, a favourite model organism of many plant biologists related to the crop plants canola, mustard, and broccoli. However, StomaAI can also be adapted by users to work on any other plant.
The AI software can accurately determine the number of open and closed stomatal pores and their density on the plant leaf in microscopy images.
The strength of StomaAI is that it can study the signalling processes that control the dynamics of stomatal movement over minutes and hours, as it has been designed to work on isolated plant epidermis to which stimuli can easily be applied.
鈥淭he real beauty of this tool is that it takes so little time to get these measurements,鈥 Dr Sai says.
鈥淯sually, people take about 100 stomata apertures at a minimum, and that can take four hours to do, whereas we can get thousands of stomata and measure finer and finer effects in a fraction of the time using this tool.鈥
鈥淚t鈥檚 not just that you can speed up the analysis, it鈥檚 that it also instantly frees you up to change the kinds of experiments you鈥檙e doing.鈥
The benefit of the study published in Plant Methods is that it provides a fast, non-destructive way to measure the stomata on intact leaves using a handheld microscope.
It can be used in the field, in greenhouses, or labs to rapidly measure stomata density and aperture. The density and pore size of stomata in wheat, rice, and tomatoes were measured by senior author Dr Abdeljalil El Habti and his team.
Dr El Habti said the tool can quickly identify plants that are more tolerant of stresses, such as droughts.
鈥淲e can now characterise populations of thousands of plants in both the greenhouse and field, providing information that can help with the breeding of plants that use water efficiently.鈥
Both studies trained a deep neural network capable of multitasking so it could translate their biological measurement issue into certain vision tasks.
The new software allows scientists to quickly measure stomata on intact plant leaves out in the field using a small handheld microscope.
Through this research, Bockman鈥攃o-first author of StomaAI鈥攕ays both groups wanted to show that it wasn鈥檛 just a machine learning problem they were looking to solve, but they wanted to test whether their creation could mimic the work of experts.
鈥淣ot only could we say it was really good at doing the computer vision task, but we were also saying it was as good as humans at doing the measuring task, which gives us assurance that it鈥檚 a drop-in replacement for a human doing it,鈥 Bockman says.
The teams are now working to consolidate the tools, with help from the , into an integrated software package.
(2023) by Na Sai, James Paul Bockman, Hao Chen, Nathan Watson-Haigh, Bo Xu, Xueying Feng, Adriane Piechatzek, Chunhua Shen, and Matthew Gilliham, has been published in New Phytologist.
鈥鈥 (2023) by Phetdalaphone Pathoumthong, Zhen Zhang, Stuart J. Roy, and Abdeljalil El Habti has been published in Plant Methods.