Written by the excellent Emilie Manny in collaboration with Université Laval in Dr. Samuel Godefroy’s research group!
Emilie Manny, Sébastien La Vieille, Silvia A. Dominguez, Gregor Kos, Virginie Barrère, Jérémie Théolier, Joseph Touma, Samuel Benrejeb Godefroy, Probabilistic risk assessment for milk in dark chocolate, cookies and other baked goods with PAL sold in Canada, Food and Chemical Toxicology (2021) 152, 112196, doi: 10.1016/j.fct.2021.112196
With collaborators in Finland, Austria and the United Kingdom I have been working on an update of the decades-old Food and Agriculture Organization (FAO) estimate that, globally 25% of agricultural commodities are contaminated with mycotoxins.
The origin of this number is very much unknown — the original report seems to have gone missing; yet the number remains widely cited in the scientific community, in articles, at conferences and in the general news media.
We took a good hard look at recent studies and results, mainly from the Joint FAO/WHO Expert Committee on Food Additives (JECFA), publicly available data from the European Food Safety Authority (EFSA) and proprietary data from the feed additive company Biomin, looking at different threshold values to establish a data-based current estimated.
Results are quite surprising and soon to be published in Annual Review of Food Science and Technology. Publication details and citation to follow once it becomes available.
Last week I gave a 2-day R workshop at the Parerea group at INAF at Université Laval. While the first day was roughly modeled on my introduction to R from my Advanced Data Analysis course at Concordia (but with agricultural data rather than meteorological data), I have added new material for the treatment of censored data and allergenic risk assessments using Monte-Carlo simulations.
The latter was also new territory for me. While I had immersed myself into allergenic risk assessments during the past 6 months for PhD student advising, implementing distribution estimations and Monte-Carlo simulations in R was a nice challenge!
Whenever I have a few hours, I work on the data analysis code for the European Food Safety Authority data on mycotoxins that I have received a few months ago.
plyr and dplyr packages (dplyr is part of tidyverse) have been instrumental to get an overview of the data, answering questions like “How many samples from country X? or How many toxins analysed in sample Y…?” Basic boxplotting of subsets (e.g., from a time range, a specific country or for a specific toxin) provide a crucial overview before conducting a detailed analysis.
I have successfully requested mycotoxin occurrence data from the European Food Safety Authority (EFSA). It’s been a lengthy, but successful process getting the data (but surprisingly transparent and being kept up to date by the legal department of EFSA), but once all member states had approved the release of their data a CD arrived in the mail a few weeks later with more than 500,000 data sets of regulated mycotoxin concentrations in a variety of raw and processes food matrices.
Together with collaboration partners from Europe, I am now in the process of analysing occurrence data for regulated toxin species. I am setting up a series of analysis scripts in R (using the tidyverse, such as dyplyr and ggplot2) to shed light on the contamination of food products with toxins such as Aflatoxins, Deoxynivalenol and Ochratoxin A.
Over the summer, two summer students have been hard at work designing and building an air sampler for volatile organic compounds. The sampler is SPME-based and analysis is performed using GC-MS. Initial testing showed promising results for ambient outdoor air.
An advanced undergraduate students will be testing and characterising the sampler as part of his final year research project!
I have co-authored a paper on portable infrared laser spectroscopy for on-site mycotoxin analysis, which nicely demonstrates future applications of spectroscopy — a combination of new powerful laser light sources (quantum cascade lasers) paired with machine learning allows for extraction of information from spectra of complex samples such as food and environmental matrices.
In this case toxin concentrations of a potent natural carcinogen (Aflatoxin B1) were used to discriminate samples at (the very low, 8 ppb) established legal limits.
Portable sensors are most useful for preliminary large scale screening on site.
My contribution on Canadian and Global emissions is part of this report: G. Kos, Y.-F. Li, D. Niemi, M. King, S.A. Smyth, C. Zdanowicz, J. Zheng, Releases of Mercury into Air and Water from Anthropogenic Activities in North America.
New research that I was involved in has been published recently — here are the articles that are now available
Y. Nazarenko, R.B. Rangel-Alvarado, G. Kos, U. Kurien, P.A. Ariya, Novel Aerosol Analysis Approach for Characterization of Nanoparticulate Matter in Snow, Environmental Science and Pollution Research (2016), doi: 10.1007/s11356-016-8199-3
This work is a reanalysis of snow samples that I have collected during my Arctic field trips with a focus on nanoparticles.
P. Kovalsky, G. Kos, K. Nährer, C. Schwab, T. Jenkins, G. Schatzmayr, M. Sulyok, R. Krska, Co-occurrence of Regulated, Masked and Emerging Mycotoxins and Secondary Metabolites in Finished Feed and Maize – an Extensive Survey, Toxins 8 (2016), 363, doi: 10.3390/toxins8120363
The manuscript provides an extensive and detailed statistical analysis of 1900+ finished feed, maize and maize silage samples from 40+ countries. 50+ fungal toxins and secondary metabolites were determined and evaluated with a focus on co-occurrence and correlation of concentrations. The analysis was performed using Matlab and R.
G. Kos, M. Sieger, D. McMullin, C. Zahradnik, T. Öner, B. Mizaikoff, R. Krska, A Novel Chemometric Classification for FTIR Spectra of Mycotoxin-contaminated Maize and Peanuts at Regulatory Limits, Food Additives and Contaminants, Part A 33 (2016), 1596-1607, doi: 10.1080/19440049.2016.1217567
As a collaborator on the EU funded Mycospec project, I have modelled mycotoxin contamination levels from mid-infrared data. Focus was the implementation of non-parametric machine learning algorithms such as bagged decision trees and a comparison with standard (and generally accepted) principal component analysis results. The analysis was performed using Matlab.