Beginners’ guide to digital phenotyping
If someone asked me what phenotyping was, I would say it was about the study of form and function: how living things appear, how they grow and what factors affect how they grow and function. That’s really all it is.
Phenotyping is something that biologists have been doing forever and of course is not restricted to plants, although for me in my job at Rothamsted Research – and for CHAP – clearly plants and crops are where the interest lies.
Essentially, phenotyping is about describing biological material, for example plants, precisely, and at a particular time and place. And while a lot of that is controlled by the underlying genes et cetera, it’s not just about the individual genes. Phenotyping is actually about the results of the action of all those genes and gene expression.
This can be measured at many different levels: from the cellular and subcellular levels, where you might be describing visual form and or chemical structure and composition – in fact anything you can measure with any instrument – all the way up to crops growing in the field and what can be described and measured at that scale.
You can measure fields of crops, and you can measure individual plants, perhaps in the field, or in greenhouses or controlled environments, and then you can even measure just parts of a plant. To measure at all these levels, you need different or adapted technologies.
There are many different aspects of the plant that can be measured in multiple ways, and different kinds of data to be captured: maybe you are measuring something specific like photosynthesis, directly, or you could be measuring growth or how tall something is, or what colour it is, or maybe you need to know the chemical composition. Any of these different physicalities are part of what is called phenotyping.
There is often an emphasis on high throughput, because these days, everything is about big numbers. People want to look at a lot of different things and want automated methods, and they want fast methods and high throughput so they can do large numbers. There is increasing emphasis on high spatial and temporal resolution in datasets, resulting in huge datasets.
All of these are often associated with what is called modern phenotyping, or digital phenotyping. However digital phenotyping is not just about high throughput and big numbers. It can sometimes be slow and painful, and determined manually. However, the key is accuracy and precision. Inevitably there is increasing emphasis on automation for high throughput.
There are lots of developments in sensors, new sensors and the use of proxy measurements for complex functions. Ideally, we would like to measure parameters of interest by simply taking photographs and extracting features from the photographs, which will inform us of important complex aspects of the subject, for example, photosynthesis in plants, which would normally have to be measured more laboriously, with dedicated instrumentation.
The images may not be simple RGB photographs but may use specific spectral wavebands or properties such as fluorescence. All of these are part of what we call digital phenotyping.
Imaging can be done by various methods: particularly successful is the use of drones. Alternatively, many solutions exploit mobile vehicles or robots with cameras attached to them. The low-tech way of doing it is to just have sensor technology that you walk around on the end of a pole.