Unlocking the power of AI and data in green life sciences

What if you could combine large sets of data on photosynthesis with AI to make predictions about the development of a crop and estimate yields more accurately? Or if we could use AI and machine learning to analyse images generated by remote sensors and drones to monitor crop growth in vulnerable areas, thus getting a better understanding of the impact of climate change.

Ten years ago, such ideas might have sounded like a distant future vision, but not anymore. Researchers in plant sciences and other green life sciences are increasingly using new technologies that allow them to generate larger amounts of reliable research data with greater ease. However, if we don’t take action the full potential of AI and data science for the green life sciences community will remain largely untapped, says Ricardo da Silva Torres, professor of Data Science and AI at WUR. ‘We can nowadays collect vast amounts of data on for instance plant development and characteristics, but without the right tools to translate this data into actionable insights, it has little value. To truly harness new technologies, stronger connections between AI, data science, and the green life sciences are needed."

Potential of data insufficiently exploited

Within WUR's recently established AI chair group, researchers are working on developing methods to achieve this. Da Silva Torres: ‘One of the challenges within science is how to better organise the huge amount of data and datasets. For example, poor annotation or noise in datasets sometimes makes them unfit for creating algorithms and data-driven research methods. More concretely, inconsistencies and inaccuracies on the annotation of gene functions, may cause AI models to learn incorrect associations, reducing their accuracy and generalisability in predictions on plant traits. As a result, opportunities are missed to develop valuable tools that can lead to relevant breakthroughs. To address this, the green life science community needs better access to data science and AI technology.’

Knowledge Discovery-BIO (KD-BIO)

One way to improve such access is to develop a knowledge infrastructure called Knowledge Discovery-BIO (KD-BIO), in which Da Silva Torres is closely involved. ‘KD-BIO is an initiative from WUR, in cooperation with six other Dutch universities and knowledge centres within the green life and health sciences and a number of tech companies. A funding application to the Dutch Research Council NWO is currently ongoing. The idea is that we will work on use cases involving AI and data science. Examples include developing improved identification methods for pathogenic fungi in food and algorithms and models that allow us to better understand and predict crop phenotypes and properties of complex ecosystems. With the aim of being able to better manage them and make them more climate-robust in the future.’

Building a strong community

According to Da Silva Torres, the uniqueness of KD-BIO is that the consortium consists of a wide range of expertise in green life sciences, data science, and AI. ‘This broad representation is essential to build and strengthen a community. To work on relevant issues using AI and learn from each other, both worlds need to know how to find each other. The domain experts and knowledge centres could address practical problems and have the required datasets, while the AI and data experts know which technologies and methods can support this. We have already set a portfolio of projects exploring AI in relevant research problems in green life sciences. PhD studies have been conducted to design AI models for unravelling climate-phenology relationships in agricultural settings and to interlink scattered data and knowledge on soil properties.”

Added value of multidisciplinary approach

That WUR is leading the KD-BIO initiative is not a coincidence, says Da Silva Torres. ‘At WUR, we not only have a lot of knowledge within all expert domains and AI, we also have a lot of experience in multidisciplinary and interdisciplinary collaboration. Issues are not approached from one expertise or only a science perspective, but always in co-creation with other disciplines and practice partners. The complexity of the challenges we face demands such a multidisciplinary approach. But it also helps advance the world of AI and data science. If practice gives a different outcome than you would expect based on the model, there might be something wrong with the design of the algorithm. The AI expert also learns from that in turn. So it works both ways.’

More impact as a researcher

During his PhD in computer science, Da Silva Torres faced for the first time the importance of interdisciplinary collaboration. ‘For my PhD, I worked on developing a search system for detecting different fish species. For this, I talked and worked a lot with biologists and fish experts. The insights I gained from this helped me a lot in creating algorithms and methods behind the system. Until my PhD, I was actually mainly concerned with the technical and computational side. That made it difficult to ultimately arrive at concrete tools that could contribute to solutions for pressing societal challenges. By pooling expertise, as we will now do at KD-BIO, my work as a researcher gets much more valuable impact.’

Path paved for further cooperation

Currently, WUR and partners are still waiting for approval of their proposal. According to Da Silva Torres, it will take until September 2025 before there is clarity. ‘We will first have to convince the NWO committee of the added value of KD-BIO. Although the funding we can get for this can give a big boost to the creation of the knowledge infrastructure, it does not otherwise affect our research agenda and the path we are on. The collaboration we have now entered into has paved the way to much more research where we can better connect AI and data science and domain knowledge. So either way, we are all moving forward together.’