(595e) A Multi-Domain Read-Across Framework to Predict Aerobic Biodegradation of Organic Chemicals

The implementation of international regulations for chemicals (e.g., REACH in the EU) requires the integrated use of a variety of testing and non-testing methods to establish the PBT profiles of new and existing chemicals. Similarity-based methods such as read-across [1,2] provide, among the variety of techniques that can be used to implement non-testing strategies, a sound and cost-effective framework to predict physicochemical properties and biological activity of chemicals. Read-across methods strongly rely on both the type of information and the similarity metric used to identify the set of reference chemicals (families) used to generate the estimate. Accordingly, given the extense variety of molecular descriptors and properties available (i.e., data domains), finding the best information to characterize chemical families for a specific endpoint is a challenging task of crucial importance.

The present work introduces a multi-domain read-across framework, based on the integration of various types of chemical information, which is applied to predict the aerobic biodegradation of organic chemicals. The data used to demonstrate this approach were extracted from the Japanese Ministry of International Trade and Industry biodegradation database [3]. Compounds with insufficient structural information or with inconsistent biodegradation measures were filtered to produce a quality-controlled database containing experimental biodegradation measures (BOD) for 1063 chemicals. The intrinsic topology preservation properties of the Self-Organizing Map (SOM) were used to identify chemical families within each data domain (e.g., constitutional descriptors, functional group counts, atom center fragments). The quality and reliability of each chemical family was quantified using consensus clustering implemented viamultiscale bootstrap methods [4]. Different approaches to compute read-across estimates (e.g., average, weighted average, value of the most similar chemical) for target  compounds (i.e., not used during SOM training) were assessed. The best results were obtained from the similarity-weighted average of the chemicals assigned to the same cluster as the target compound. Finally, consensus techniques [5] were used to combine, in an optimal way, the output of each individual SOM to yield a final and more accurate read-across estimate. Results obtained using the new multi-domain read-across method outperformed results obtained via conventional read-across based on a particular type of information.

This work demonstrated that multi-domain read-across provides a suitable framework which is able to identify complementary chemical families with similarity features that remain hidden when only data from a single domain are used. Ongoing research work aims to include mechanistic information to enhance the characterization of the chemical families with the inclusion of biological information for the biodegrading organisms (e.g., activity of genes involved in biodegradation pathways).

References

[1] Schuurmann G., Ralf-Uwe E., Ralph K. (2011) Quantitative read-across for predicting the acute fish toxicity of organic compounds. Environ. Sci. Technol.  45, 4616–4622.

[2] Enoch SJ, Cronin MT, Schultz TW, Madden JC, (2008) Quantitative and mechanistic read across for predicting the skin sensitization potential of alkenes acting via Michael addition. Chem Res Toxicol, 21(2):513-20.

[3] CITI - Biodegradation and Bioaccumulation data of existing chemicals based on the CSCL Japan (1992). Chemicals Inspection and Testing Institute, Japan, ISBN 4-98074-101-1.

[4] Rallo R., France B., Liu R., Nair S., George S., Damoiseaux R., Giralt F., Nel A., Bradley K., Cohen Y. (2011) Self-Organizing Map Analysis of Toxicity-Related Cell Signaling Pathways for Metal and Metal Oxide Nanoparticles. Environmental Science & Technology. 634-640.

[5] Fernández A; Lombardo A; Rallo R; Roncaglioni A; Giralt F; Benfenati E. (2012) Quantitative consensus of bioaccumulation models for integrated testing strategies. Environment International, (in press, doi: 10.1016/j.envint.2012.03.004).

 Acknowledgement. This research was financially supported by the Spanish Ministry of Economy and Competitivity (CTM2011-24303).