Plants adapt to their environment, but through what mechanisms? In an 'Insight' review in Nature, Wim van der Putten (NIOO-KNAW) and Ben Scheres (WUR) enlist concepts from artificial intelligence to open up new perspectives on plant growth and development.

Ecological analyses of how plants cope with their environment tend to be limited because "they treat the trade-off mechanisms that underlie plant responses as a black box", write the two scientists in Nature.

Going into that black box are external factors such as the quality and quantity of light, the availability of nutrients, drought or toxicity stresses, physical damage and pressure from pathogenic agents.

Coming out, there is a wide variety of growth forms and patterns of life events. But what exactly happens in between?

Gene to Gaia

Scientists who study plant development approach this question from opposite sides of the spectrum, write Scheres and Van der Putten in their 'Insight' review.

Plant biologists "build upwards from the molecular and physiological level to understand organ and whole-plant responses in a limited set of well-defined model species", while ecologists "deduce mechanisms from the ecology of species and their interactions in specific, multispecies environments".

To bridge the gap between these contrasting approaches - between gene and Gaia - Scheres and Van der Putten introduce a concept borrowed from machine learning: that of the perceptron.

Neural networks

A perceptron is a simple neural network that consists of one or several interconnected layers. Processors - artificial neurons - are linked to other processors, and those links can be strengthened, weakened and severed, or entirely new links formed.

In artificial intelligence, perceptrons are used for supervised learning of 'binary classifiers', i.e. functions that can decide whether an input, represented by a vector of numbers, belongs to some specific class or not.

In other words: they can be trained to recognise simple patterns, as in optical character recognition (OCR). The more input is processed through the network's interconnected layers, the more effective it is in making choices.

Limited variation

OCR is a good example, because regardless of the input there are only 26 characters in the alphabet to choose from. Scheres and Van der Putten write in Nature that something similar applies to plant development.

Although they can grow under the most diverse and extreme conditions, "most plant species can be characterized by variations in only a few important trait characteristics, of which only certain combinations are well represented in nature."

Considering diversity in plant responses to the environment as the adaptation of an information-processing neural network, could therefore make it easier to analyse non-model species under natural conditions. The novel approaches this produces, conclude the two authors, could well be used for enhancing the sustainability of future world food production.