|Abstract Title:||Association between Hg prenatal exposure and neurodevelopment: Application of multi-target prediction methods|
|Presenter Name:||Janja Snoj Tratnik|
|Company/Institution:||Jožef Stefan Institute|
|Session:||Progress in understanding Hg and human health impacts|
|Day and Session:||Tuesday 26th July - Session Two|
|Start Time:||10:00 UTC|
|Co-Authors:||Janja Snoj Tratnik,Stefan Popov,Martin Breskvar,Darja Mazej,Milena Horvat,David Neubauer,Zdravko Špirić,Igor Prpić,Sašo Džeroski|
Abstract Information :
We have applied machine learning algorithms with an aim to explore association(s) between Hg, micronutrients, socio-economic, health- or life-style related features in prenatal, early life and childhood and neuropsychological development.
The following data has been used from the existing PHIME longitudinal birth cohort study including mother-child pairs from Slovenia and Croatia: life-style; concentration of elements in biological samples; Bayley-III scores (18 months) and WISC-IV scores (7-8 years of age). We have applied single- and multi-target (semi-) supervised machine learning methods to build predictive models for the Bayley-III and WISC-IV scores. Additionally, we have used feature ranking methods to estimate the importance of individual attributes on the predicted scores.
The supervised Predictive Clustering Tress (PCT) model identified the child?s gender, the concentration of methyl Hg in the mother?s blood and the mother?s age as the most relevant attributes for prediction of multiple Bayley scores ? cognitive, language and motor - simultaneously. The semi-supervised PCT model (utilization of missing data) identified the concentration of methyl Hg in the cord blood and the number of pregnancies as most relevant attributes. In the single-target approach, total Hg in mother?s milk was identified as the most important attribute for prediction of fine motor score. In a subpopulation of children who were followed at 7-8 years of age, some additional features have been uncovered that seem to condition the response. Among those, the most evident are Zn with beneficial effect at higher levels and Cu with the opposite effect, certain socio-economic and body weight related attributes.
The use of novel multi-target prediction methods confirmed the existence of negative association between prenatal or early postnatal Hg exposure and fine motor development, as shown in our previous studies, and revealed some other factors that seem to compensate the adverse effects of Hg at low levels of exposure.