# A Statistical Exploration of QSAR Models in Cancer Risk Assessment: A Case Study on Pesticide-Active Substances and Metabolites

## TL;DR

This paper explores how QSAR models can help assess cancer risk from pesticides, highlighting their potential and limitations.

## Contribution

The study provides a methodological analysis of QSAR model coherence in carcinogenicity assessments of pesticides.

## Key findings

- QSAR models show significant potential for cancer risk assessment of pesticide substances.
- Inconsistencies across models and applicability domain constraints were identified as key limitations.
- Transparent definitions of model applicability domains are critical for reliable results.

## Abstract

Data generated using new approach methodologies (NAMs), including in silico, in vitro, and in chemico approaches, are increasingly important for the hazard identification of chemicals. Among NAMs, (quantitative) structure–activity relationship (Q)SAR models occupy a peculiar position by allowing (in principle) a toxicity estimate on the sole basis of chemical structural information, leveraging upon toxicity profiles of already tested chemicals (a training set). Consequently, the metrics adopted for the estimation of both the congruence of the test chemicals with the training set and the risk categorization are of paramount importance. This paper comprises a small-scale, mainly methodological study to investigate these aspects and assess the general coherence between the results from different (Q)SAR models applied to the assessment of the carcinogenicity of pesticide-active substances and metabolites. The results of the present study underline the significant potential of using (Q)SAR models, together with limitations, such as inconsistencies in results across models and the intrinsic constraints of their applicability domain. The critical role of a priori strategies adopted in defining the applicability domain of the models is highlighted, emphasizing the need for user-transparent definitions. This is a crucial step for a sensible integration of the information coming from different NAMs.