Spectroscopy is the study of the absorption and emission of light and other radiation by matter. It is used in a wide range of scientific fields and is particularly useful in biology. Spectroscopy provides data on the presence of proteins in blood, for example, or fats in dairy products.

Blood or milk samples are assessed using methods such as mid-infrared (MIR) spectroscopy. A sample’s MIR spectrum gives clinicians or researchers data that can be used to predict the quantity of, for example, fat molecules in a sample of milk.

However, researchers cannot predict exactly the quantity of fat molecules from the MIR spectrum as there is always uncertainty in the reading. The MIR spectrum suggests an amount of fat present in the sample but how close is that value to the true amount of fat? How accurate – or not – is the prediction from the MIR sample? Can we quantify the uncertainty in the prediction?

The information in the uncertainty matters. In the case of fat in milk, food manufacturers creating infant formula or cheese need to know about these properties of a milk sample in order to decide how to process it. Predicting a range of fat content, based on the predicted value and the associated uncertainty, will enable more informed decisions than basing a decision on a single value.

That is where the work of Insight Principal Investigator Professor Claire Gormley comes in. Prof. Gormley uses a statistical modelling framework known as Bayesian inference to add more value to predicted values by also delivering an estimate of their inherent uncertainty. The work builds on a popular algorithm for prediction from MIR spectra, called partial-least squares regression, but uses the famous Bayes theorem to ensure accurate predictions alongside a measure of the uncertainty. 

Prof. Gormley and her team have produced open-source software that can be downloaded by anyone using MIR spectroscopy or similar data to predict multiple traits simultaneously and quantify their uncertainty. The open-source R package (called bplsr), was published on the R statistical computing platform in November 2024 and has had 1750+ downloads since. This project was in collaboration with Dr Szymon Urbas (now in Maynooth University) and researchers from the VistaMilk Reseaerch Ireland research centre. The work built on foundational work carried out by Insight researchers in Insight 1 and 2, by former Insight funded PhD student Dr Keefe Murphy (now in Maynooth University) and Insight funded postdoctoral researcher Dr Suzy Whoriskey (now in University of Strathclyde). 

Prof. Claire Gormley graduated with a PhD in Statistics from Trinity College Dublin in 2007. During her doctoral studies she spent five months as a Visiting Scholar in the Department of Statistics/Centre for Statistics and the Social Sciences in the University of Washington Seattle, USA. She joined UCD in September 2006, and has been Full Professor of Statistics since 2020. She was appointed a Principal Investigator in Insight in 2025.

Link to paper: doi:10.1214/24-AOAS1947