MRC DiMeN Doctoral Training Partnership: Use of machine learning methods to develop new biomarkers for nutrition induced toxicity

The safe intake level for vitamin A is currently not well defined, since the lowest-observed-adverse-effect level (LOAEL) is set differently by the US Food and Nutrition Board and the European Scientific Committee on Food. Data are urgently needed to define an appropriate upper limit of safe intake for children, to enable accurate assessment of the risk of excess vitamin A intake from overlapping international programs.

Concerns about inadvertent chronic excessive retinol intakes due to overly frequent supplementation combined with concurrent use of fortified staple foods, micronutrient powders and voluntarily fortified commercial products have been raised (Kraemer et al. 2008, Allen and Haskell 2002). Chronic excessive VA intake can create liver abnormalities (Allen and Haskell, 2002; Stickel et al. 2011) and increase the risk of bone fracture in Western countries (Penniston and Tanumihardjo, 2006). However, the link between a higher incidence of fractures, lower bone mineral density (BMD) and higher vitamin A intakes remains speculative due to methodological issues around the accurate assessment of vitamin A intake, status and adequate biomarkers.

Our groups have undertaken a range of studies funded by the Bill and Melinda Gates Foundation that involved isotope dilution techniques combined with a range of OMICS technologies in the search for new biomarkers that would allow us to define the new upper limits of intake for this vitamin. However, the analysis of the metabolome alone indicated a range of over 1000 potential markers that now need validation. To allow us to integrate the results from metabolome, proteome and microarray analysis, this project aims to use machine learning methods to (a) extract identify reduced panels of biomarkers from combined multi-omics data (Swan et al., BMC Genomics 16(S1):S2, 2015; Lazzarini et al., BMC Bioinformatics 18:322, 2017Lazzarini et al., Osteoarthritis and Cartilage, 2017 25(12):2014-2021)) and (b) infer or functional networks (Lazzarini et al., BioData Mining 2016, 9:28) from these on-going experiments.

The project will verify this new approach of data mining by investigating the synergistic effects of retinol stores and aflatoxin exposure on liver toxicity in samples obtained from both tissue culture and animal experiments. Retinoic acid, the metabolically active form of vitamin A, can increase the toxic effect of aflatoxin due to transcriptional modifications of Cytochrome P450. Since currently 500 million people worldwide are exposed to aflatoxins in the diet at levels which increase both morbidity and mortality, and since the consumption of aflatoxin will become more prominent over the coming years due to global warming, new upper safe limits of intake are particular important in low and medium income countries (LMIC), such as Guatemala, where consumption of both Aflatoxin (due to ideal environmental condition for growth on crops) and dietary retinol (due to large scale food fortification) are higher than the current safe intake levels.

The integration of traditional OMICS technologies with machine learning methods will allow this project to set new guidelines in how best to set upper safe limits of intake for both nutrients and environmental agents that interact with the nutritional status of the host to establish health or disease.