Why interdisciplinary science is needed more than ever

The lines between separate scientific disciplines are becoming more blurred. Solving today’s problems often requires teams of scientists from a range of specialisms. But multidisciplinary collaboration also has challenges, in particular the need to “speak the same language”, ask the “right” questions and be familiar with techniques and knowledge that exist in other fields.

To see the importance of finding a common language look no further than the rapid uptake of large language models (LLMs) such as ChatGPT. LLMs can be convenient research aids, but the information provided by them is not always accurate. We can ask LLMs questions about another field, but without existing domain knowledge we cannot always tell if the answers are reliable.

Getting up to speed with a new research field can be tricky – it’s difficult to understand everything fully, but tempting to think that you do. There’s a parallel with sport where it might sound reasonable, say, to assume that mixed martial arts (MMA) fighters can easily become boxers. However, the evidence suggests that MMA fighters often struggle against professional boxers even though fist fighting uses a subset of the skills needed to be successful in MMA.

Back in academia, it’s common to get pushback from “real experts” whenever grant proposals or papers drift too far outside one’s own comfort zone. Nevertheless, discipline mixing is needed more than ever. Today’s problems often straddle different scientific disciplines: how to treat large, complex datasets, for example, is a common challenge in many different fields.

Look up at the stars and not (just) down at your tea

We realized this recently in our work at Queen’s University Belfast, which has been pushing for researchers to share their data analysis strategies with colleagues in other fields. In our case, we had been collaborating with Yicong Li at the Institute for Global Food Security on infrared and ultraviolet-visible spectroscopy and machine-learning models for monitoring the freshness of fish, which required only a few samples for analysis.

However, many food studies need hundreds or thousands of samples to be analysed and class imbalances can quickly arise in which some types of foodstuff have more examples than others. This can then lead to training datasets that do not produce predictive models. One example is tea, which Li has been investigating recently, again via spectroscopy and machine learning, using many samples from all over the world.

Li was trying oversampling, which creates synthetic data to equalize class imbalances. Yet over in the Queen’s physics department, we discovered another strategy was being used to classify problems in astrophysics. Matt Nicholl and PhD student Xinyue Sheng had been working on predicting the classes of energetic cosmic explosions, based on an image of the galaxy where they occurred. They wanted to train their model to find particularly rare classes, so their training set had the same problem: there were only a handful of examples of some classes of interest.

In addition to oversampling, they were also using a “weighted loss function” in their training, in which weights were inversely proportional to the number of examples in a given class. Their approach led to a substantial improvement in their astrophysics application, but it turns out the basic idea is completely general in nature and can be just as easily applied to tea.

Sleeping beauties

Knowledge exchange does not only concern data, but sometimes a whole set of ideas. An interesting study of citation metrics in 2015 by researchers at Indiana University found that there is a class of papers that receive very little attention for years before suddenly shooting skywards with a deluge of citations. Notably, these “sleeping beauty” papers include Albert Einstein, Boris Podolsky and Nathan Rosen’s work in 1935 examining non-locality in quantum mechanics, which led to John Bell’s theorem in 1964 and ignited significant interest in the original “EPR” paper.

Such citation trends can arise because the papers’ findings are adopted by researchers in a different field. Other similar instances include work in the 1930s and 1940s on hydrophobic theory, which describes how certain substances minimise their contact with water. Yet perhaps the sleepiest of sleeping beauties is the principal component analysis (PCA) work by Karl Pearson, which slumbered for over 100 years before “awakening” in the early 2000s.

PCA – a technique that simplifies complex datasets by reducing the number of variables while minimizing information loss – had already been gaining traction during the 1980s and 1990s when matrix calculations became easy for computers alongside the development of statistical software packages and open scripting environments. In research papers published today it would be unusual not to see PCA used as an exploratory tool for multivariate dataset analysis.

As these examples show, it’s crucial that communication channels are open between varying fields. However, too many academic researchers can get siloed. Interdisciplinary science hubs are one way to break down barriers, acting as spaces to exchange ideas between scientists.

One example that we have been involved with is Smart Nano NI, which is a consortium of universities and photonics-based companies in Northern Ireland. It recently released TITAN, a bio-process analysis system based on gold nanostructured chips, for real-time bio-analysis. Smart Nano NI is now moving from benchtop to backpocket, looking to develop fully miniaturized sensing devices by integrating different kinds of photonic components like lasers, filters and detectors, all on the same chip.

Elsewhere, centres for doctoral training – such as the Photonic Integration and Advanced Data Storage programme with the University of Glasgow – bring together groups of PhD students to work on various projects under a common theme. These schemes not only foster new ideas with the student cohort but bring together academics to bridge different parts of research. Either way, we are getting people talking and interested in emerging scientific questions.

So if you are sitting on a problem, there might be a chance that someone in a different field has solved it or at least offered the tools to do so. As our sky-gazing friends might say, “There is nothing new under the Sun.”