Predicting Early Life Stage Toxicity of Fish Using Intelligent Consensus Based Approach: Data Tested in Compliance with OECD Test Guideline 210 

Souvik Pore^a,$, Alexia Pelloux^b, Anders Bergqvist^b, Mainak Chatterjee^a, Kunal Roy^a, *

^aDrug Theoretics and Cheminformatics Laboratory, Department of Pharmaceutical Technology, Jadavpur University, 188 Raja S C Mullick Road, 700032, Kolkata, India

^bGlobal Product Compliance (Europe) AB, Ideon Beta 5, Scheelevägen 17, 223 63, Lund, Sweden 

Abstract 

Early life stage (ELS) toxicity testing in fish plays a crucial role in understanding the long-term effects of various chemicals, including pesticides, industrial compounds, pharmaceuticals, food additives, etc. This testing methodology is particularly important for assessing and prioritizing a wide range of chemicals under the Registration, Evaluation, Authorization, and Restriction of Chemicals (REACH) regulation, which aims to ensure the safe use and management of environmental substances. To overcome the limitations of traditional testing—such as high costs, huge time, and resource consumption—in silico methods can be harnessed to predict the toxicity of chemicals for which experimental data is unavailable. In this research work, we have developed a predictive Quantitative Structure-Activity Relationship (QSAR) model to assess the chronic effects of numerous chemicals on the early life stages of fish. To build the models, we gathered toxicity data from two sources: J-check and eChemPortal. These databases provide comprehensive summaries of experimental studies that adhere to the guidelines outlined by the Organization for Economic Co-operation and Development (OECD) Test Guideline 210. The collated data include two types of endpoints: the No Observed Effect Concentration (NOEC) and the Lowest Observed Effect Concentration (LOEC), both of which were used to construct the QSAR models. Initially, we developed six distinct partial least squares (PLS) models, each employing different combinations of chemical descriptors to analyze both endpoints. These models were used to generate consensus-based predictions to enhance the predictive accuracy for unknown chemicals. Among the developed models, consensus model-3 and individual model-3 showed promising performance metrics of Q²F1=0.71 & Q²F2=0.71, and Q²F1=0.80 & Q²F2=0.79, respectively. To ensure the reliability and 

accuracy of our models, we validated them against experimental data of nine industrial chemicals provided by Global Product Compliance (Europe) AB. Finally, the validated QSAR models were utilized in a systematic screening and prioritization process for chemicals collected from the Pesticide Properties Database (PPDB) and DrugBank databases to identify those with potential long-term impacts on fish and the broader ecosystem. 

Keywords: Fish early life stage toxicity; Quantitative structure-activity relationship; OECD  210; No observed effect concentration; Lowest observed effect concentration; Partial least squares; Intelligent consensus-based prediction; Pesticide Properties Database; DrugBank.