A Data-Driven Approach for Improved Effective Classification in Predictive Toxicology

Effective multi-class classification for complex data in real-life problems is an open-ended challenge. Prediction of toxic effects of chemical compounds based on experiments involving animals and human beings is very expensive in terms of time, social and financial cost. Therefore it is vital to make use of all available information obtained from experiments and build up a more effective hybrid classification/prediction system to incorporate any available useful piece of knowledge for initial in silico toxicity validation. The paper proposes a correlative data-oriented fusion algorithm to develop effective models based on multi- source data for classification applied to predictive toxicology. Traditionally hybrid intelligent systems integrate just models build on individual data sets by some sort of voting algorithms. We propose an algorithm to generate improved classifiers by use of correlative information of chemical compounds on different endpoints for effective classification.