Concentration-dependent response relationships provide essential information on the characteristics of chemicalinduced effects on toxicological end points, which include effect (inhibition or induction), potency, and efficacy of the chemical. Recent developments in systems biology and high throughput technologies have allowed simultaneous examination of many chemicals at multiple end point levels. While this increase in the quantity of information generated offers great potential, it also poses a significant challenge to environmental scientists to efficiently manage and interpret these large data sets. Here we present a novel method, ToxClust, that allows clustering of chemicals on the basis of concentrationresponse data derived with single or multiple end points. This method utilizes a least distance-searching algorithm (LDSA) to measure the pattern dissimilarity of concentrationresponse curves between chemicals and their relative toxic potency. ToxClust was tested using simulated data and chemical test data collected from the human H295R cell-based in vitro steroidogenesis assay. ToxClust effectively identified similar patterns of simulated data and responses to the exposure with the five model chemicals and separated them into different groups on the basis of their dissimilarities. These observations demonstrate that ToxClust not only provides an effective data analysis and visualization tool, but also has value in hypothesis generation and mechanism-based chemical classification.
IntroductionThere is urgent need for methods that allow evaluation of large numbers chemicals for their potential toxicity and to prioritize these chemicals for further testing (1). Recent developments in toxicogenomics and high-throughput screening techniques, including among others those broadly defined as genomics, transcriptomics, proteomics, and metabonomics utilizing multiple molecular and cellular end points has enabled testing of large numbers of chemicals (2). These new techniques are increasingly being used in priority chemical screening programs such as Tier 1 of the Endocrine Disruptor Screening Program of U.S. Environmental Protection Agency (EPA) (1). During these screening initiatives, researchers collect large amounts of multidimensional data (e.g., gene transcripts, proteins, and enzyme products) for various concentrations of each chemical analyzed (2). One of the major remaining challenges is the prioritization of chemicals for further testing based on the information obtained during Tier I (3). However, because of the lack of appropriate data evaluation techniques that effectively handle large quantities of diverse data, progress in the classification of chemicals by mode-of-action has been slow and is not yet part of many regulatory programs entrusted with this action (4). To accurately classify chemicals based on their critical mechanism of action so that they can be grouped for risk assessments (2) and to take advantage of detailed information on a few model chemicals, we must know the mechanism of action. How...