Construction of Hybrid models by fusing the Read-Across concept with the QSAR framework for assessing Developmental and Reproductive toxicity (DART) tested under OECD TG 414 

Sapna Kumari Pandey^{1, $} and Kunal Roy^{1, *} 

¹Drug Theoretics and Cheminformatics Laboratory, Department of Pharmaceutical Technology, Jadavpur University, Kolkata 700032, India 

Abstract 

Developmental and reproductive toxicity (DART) are relatively poorly described endpoints in the toxicological studies of humans and other organisms. The expense and animal participation are prohibitively high for doing DART testing on a single molecule and comprehending the various aspects of DART estimation. The application of alternative methodologies to foresee this type of toxicity is thus attracting a lot of interest. In this case study, a high-quality DART database was collected from the Integrated Chemical Environment (ICE) comprising chronic toxicity data i.e., Lowest Observed Effect Level (LOEL) data of pharmaceutical medicines and excipients, agriculturally significant chemicals, and a few miscellaneous categories of molecules. Novel In-silico hybrid models were created in this study for the adult and fetal stages of rodents (rats and mice) and rabbit life. For a more precise study, we have manually segregated the data based on species (rodents and rabbits) and life stages (adult and fetal). Therefore 4 hybrid PLS models were developed for four datasets—adult rodent (Dataset 1), fetal rodent (Dataset 2), adult non rodent/rabbit (Dataset 3), and fetal non-rodent/rabbit (Dataset 4) in this study. The models were developed by combining the attributes of the developed classical QSAR (Quantitative Structure-Activity Relationship) model with similarities-derived features (obtained using the quantitative read-across (qRA) predictions). The primary objective of our study was to check the statistical quality of our developed hybrid models with the traditional QSAR models along with ease of interpretation and transferability for addressing complex endpoints like DART testing by using the knowledge of molecular features and similarity-based features. All 4 hybrid models' internal and external measures outperform the traditional QSAR and qRA predictions ranging from 0.66 to 0.76 for R², 0.58 to 0.66 for Q²LOO, and 0.57 to 0.67 for Q²F1. The predictivity and transferability of the model are enhanced by the new hybrid methodology compared to traditional QSAR and qRA hypotheses, which may be used as a substitute in a regulatory setting to close data gaps in the evaluation of DART testing. From the developed models for rodents and rabbits, we have identified that the complexity of the structure, lipophilicity, the presence of oxo or thiophosphates, and intermolecular hydrogen bonding with the water molecules are the core molecular features responsible for DART toxicity. This method can also be used with other testing methods to create a comprehensive plan for reviewing and examining existing or new chemicals for DART testing, as it is not feasible to investigate all aspects of the ever-increasing number of chemicals by experimental studies.

Keywords: Developmental and Reproductive toxicity; Integrated Chemical Environment; Lowest Observed Effect Level data; In-silico hybrid model, Quantitative Structure-Activity Relationship; quantitative read-across;