Boosted Regression Trees: A forgotten tool in ecotoxicology multivariate data understanding? Their potential use in biomarkers selection.

BERTRAND, L.; MONFERRÁ, MAGDALENA VICTORIA; AMÉ, MARÍA VALERIA

Petnica

Congreso; 4th YOUNG ENVIROMENTAL SCIENTIST MEETING; 2015

Boosted regression trees (BRT) seek to identify a single ?best? model describing relationships between the response (Y) and the predictor variables (Xs). Boosting progressively it builds a sequence of models of increasing complexity, each one fitting the training data slightly better than its predecessor. BRT has advantages over other statistical methods as they are robust to variable collinearity, variable outliers, and missing data, and can include both categorical and continuous variables. Moreover, as ecotoxicology moves towards more comprehensive ecological risk assessment procedures, better analytical tools are needed, particularly for analysis of large, complex datasets with many predictor variables. While their use has become increasingly popular in other fields, especially ecology, a poor use has been given in ecotoxicology. Our aim was to select those biomarkers with higher capacity to predict Zinc exposure concentration of Palaemonetes argentinus and their posterior use in an integrated biomarker index (IBR), as a potential biomotoring tool. Shrimp have been exposed at control and three environmental zinc concentrations (5, 50 and 500 µg L) during 96 h. After that, 17 biomarkers (Bm) including enzymes activity, effect and exposure Bm, have been measured in cephalothorax and abdomen sections. BRTs have been carried out with those Bm that showed a significant response at experimental conditions tested when compared to control condition. Models were constructed with the GBM packages in R software and PRESS function was applied to confirm the capacity of each biomarker selected to predict the toxic exposure concentration. Finally, Bm selected with BRTs were introduced in IBR function. According with our results four Bm (soluble zinc, metallothioneins, acetyl-cholinesterase activity and insoluble zinc) have been selected in cephalothorax and used to build an IBR. A strong correlation (R2= 0,98) was found between IBR values and exposure concentrations suggesting that the IBR might be a useful tool for quantification of integral damages induced by Zinc. By the other hand, in abdomen, only two Bm were selected by BRT, which didn?t allow us to build an IBR. Results stand out the greater potential of use BRT as a Bm selection?s tool. Finally, Bm measured in cephalothorax of P. argentinus showed higher capacity to explain metal exposure than those determined in abdomen and are pointed out as suitable to future use in biomonitoring studies.