BioSS staff have collaborated with Rowett scientists in studying recently collected data to look for associations between mycotoxin levels and micronutrient and heavy metal levels in samples of Scottish cereals.  Mycotoxins are toxic chemicals produced by fungi, typically those infesting crops.  Micronutrients are essential elements of nutrition, albeit in small quantities, whereas some heavy metals can be harmful.  Hence, the presence or absence of these all have implications for health.  Identifying evidence for positive and negative relationships between cereal micronutrient / heavy metal and mycotoxin levels would have opened up avenues of research to reduce mycotoxin burden and improve food safety within Scottish cereals.  However, detailed statistical analysis indicated that apparent associations were most probably spurious, and certainly did not provide strong evidence for genuine relationships.

The collected multivariate data set consisted of data detailing the quantity of 16 micronutrients or heavy metals and multiple mycotoxins in 125 samples of cereals grown in Scotland.  The key objective was to identify any statistically significant association between any of these two types of variables.  Such associations can easily be quantified using standard correlations, but testing these for statistical significance gives rise to a number of difficulties.  There are large numbers of correlations being calculated, and so, purely by chance, some can be expected to be close to ±1, and hence to have a statistical p-value less than the 5% level conventionally taken to indicate evidence for the presence of a real pattern or effect.  An additional complication was that there were subgroups of year, region and season in the data collected; looking at relationships in these subgroups further increased the risk of reporting false positives.  Simple, traditional, adjustments such as Bonferroni corrections are unsuitable for data with such structures.  In addition, there are correlations within the sets of micronutrients / heavy metals and mycotoxins (these were expected and confirmed), which provide an additional layer of complexity.  BioSS researchers calculated many correlations, and had subsequent discussions with Rowett scientists on their interpretation.  The uses of principal component analysis and partial least squares analysis were explored to see if it was possible to detect associations that were more complex than just simple correlations between individual micronutrients and individual mycotoxins.

Although potentially disappointing to collaborating scientists and policy makers, the final conclusion of the full analysis was that there was no evidence that meaningful associations between levels of micronutrients / heavy metals and mycotoxins were present.  It is important to use appropriate statistical methods in data analysis, reducing the risk of reporting spurious positive results and thus contributing to the reproducibility crisis in science.  

This work was done in collaboration with Alice Sneddon at The Rowett Institute and was funded under the Scottish Government's Strategic Research Programme for environment, agriculture and food.