Among scientific disciplines, botany might be considered one of the least tech-minded branches, concerned as it is with the natural world of plant life. But like the rest of biology, botany is quickly moving into the types of large-scale experiments that require more sophisticated and advanced techniques. In many botany labs, high-throughput sequencers generate data at unprecedented rates about plant genomics for many different species. However, this genetic bounty creates a new bottleneck, as the complementary studies examining how those genes control plant traits still proceed at a speed closer to that of old-fashioned fieldwork. That old cliche "watching the grass grow" is not compatible with fast-paced science.
To help bring phenotype closer to the pace of genotype, Nicola Ferrier has equipped botanists with powerful new lab assistants: robots. Ferrier, now an engineer at Argonne National Laboratory, worked with University of Wisconsin botanists to design better equipment for monitoring the growth of the plant species Arabidopsis thaliana. Arabidopsis is a small flowering plant popular as a laboratory model species, in part because of its relatively small genome of roughly 27,000 genes. Eventually, scientists would like to find the role of each of those genes on aspects of the plant's phenotype, such as root gravitropism, how the roots grow in response to gravity.
But the popular method – a computer-controlled camera to monitor the growth of one Arabidopsis seedling at a time – was far too slow to monitor tens of thousands of mutants. So Ferrier helped the laboratory of Edgar Spalding replace their single-camera system with a "robotic machine vision platform" capable of monitoring up to 144 seedlings simultaneously as their gravity is artificially changed (by rotating the dish 90 degrees).
The setup combined a robot-controlled camera with a fixture that holds 36 petri dishes, each containing four specimens, and over 100 lights for both growing and filming the seedlings.
Putting together this sophisticated machinery was only the start. As the camera moved to film each dish, it had to find each root, get it in focus, take an image and move on to the next one – over and over again, cycling through each of the 144 specimens. Many times, the images were far from perfect, with motion blurs, out-of-focus shots and condensation on the surface of the dish contributing to subpar results. But after writing a computer vision algorithm training the camera to properly identify the location of the root – looking for dark, skinny, elongated objects, Ferrier said – performance was improved to successfully acquire usable data from 87% of the samples in a given run, increasing the rate of data collection more than a hundred-fold.
Designing robots to monitor plant growth was about much more than just relieving tedium for human researchers, Ferrier said. The system allowed scientists to collect information about hudnreds of traits at hundreds of time points from thousands of seedlings grown in the same conditions, a much richer dataset in terms of time and more controlled dataset in terms of environment. In turn, the deeper pool of data created opportunities (and challenges) for computational analysis – one early experiment required 70 CPU years to create a single quantitative trait locus plot.
"Together, these things may change the way we think about phenotypes," Ferrier said. "We've really leveled the playing field in terms of tools for studying genomes and tools for studying phenotypes."
Extending the work, Ferrier is currently helping University of Chicago evolutionary biologist Joy Bergelson adapt the robotic monitoring system from the laboratory to field studies of Arabidopsis, and collaborating with other scientists to develop new technologies for studying maize. With Computation Institute fellow Kate Keahey, Ferrier is also starting a new field project to help create monitoring and computational modeling tools for small community farms, giving those farmers some of the same technology that larger farming operations use. Someday soon, robots may be important contributors to botany and agriculture, adding a science-fiction twist to traditionally rustic fields.