Nanotoxicity modeling in multidimentional cube

Nanotoxicity modeling can reveal the relationship between nanomaterial properties and unintended adverse effects. Traditional modeling usually builds a prediction model on the whole dataset. It does not examine the subsets of the data and their quality for model building. In this paper, we introduce a prediction cube approach to nanotoxicity modeling. Prediction cube is a new type of data cube for data exploration and predictive analytics. It can help researchers slice/dice a large amount of nanotoxicity data and measure the quality of different subsets for building prediction models. We constructed a prediction cube using a sample nanotoxicity data on zebrafish. The results show that the prediction cube can help identify useful subsets for building high quality prediction models. And the cube interface facilitates the exploration of data subsets and associated models.