Citation:

Tornero-Velez, R. Using Big Data Analytics to Address Mixtures Exposure. ICCA-LRI Workshop 2017: “Fit-For-Purpose Exposure Assessments for Risk-Based Decision Making”, Como, ITALY, June 21 - 22, 2017.

Impact/Purpose:

Unsupervised machine learning approaches for assessing co-occurrence patterns of exposure. These approaches are relevant for assessing chemical mixtures.

Description:

The assessment of chemical mixtures is a complex issue for regulators and health scientists. We propose that assessing chemical co-occurrence patterns and prevalence rates is a relatively simple yet powerful approach in characterizing environmental mixtures and mixtures exposures in the population. We first describe an ecological model for community assembly of avian species and draw a parallel with the co-occurrence of environmental chemicals in that both are driven by processes that introduce structure in observed patterns of co-occurrence. We then show that techniques used for assessing species co-occurrence patterns in community ecology are similar to techniques used for discovering relationships among commercial transactions for marketing purposes. Specifically, we show that the site x species presence-absence (0,1) matrix from community ecology is representable as a commercial transaction database for unsupervised machine learning approaches employed in analyzing high volume purchasing data. These techniques, which include frequent itemset mining (FIM) and association rule mining (ARM), are well suited for addressing chemical co-occurrence because they return a measure of support (prevalence) for combinations of interest and provide measures of association among the chemical actors. Application of these mining tools is shown for: (1) identifying prevalent combinations of biomarkers in the population; (2) revealing co-occurrence patterns in consumer product purchases; and (3) revealing relationships among chemical and non-chemical stressors. Given the availability of site x species presence-absence data and these informatics tools, the daunting task of addressing mixtures exposures is more approachable. In turn, these approaches can enable more effective cross-discipline (toxicology, exposure and risk assessment) strategies to address mixtures.

URLs/Downloads:

Record Details: