Comparison of Cramer classification between Toxtree, the OECD QSAR Toolbox and expert judgment

Establishment of a threshold of toxicological concern for pharmaceutical intermediates based on historical repeat-dose data and its application in setting health based exposure limits

Availability of toxicological data for pharmaceutical intermediates (IMs) used in the manufacture of small molecules is often limited. Scarcity of data - in particular, repeat-dose toxicity (RDT) - renders the calculation of health-based exposure limits (HBELs) problematic. Establishment of HBELs, including occupational exposure limits (OELs) and permitted daily exposures (PDEs) facilitating worker and patient safety respectively, is however essential. Historic 28-day oral rodent toxicity data was analysed for 103 GSK isolated IMs. No-observed (adverse) effect levels (NO(A)ELs) and critical effects were extracted. The 5th percentile (p05) of the NO(A)EL distribution was 15 mg/kg/day. Substance specific HBELs were calculated, selecting the NO(A)EL as the Point of Departure (PoD); 99% of IMs (n = 102) were assigned an oral PDE ≥1000 μg/day and OEL ≥100 μg/m3. A default oral PDE of 1000 μg/day and OEL of 100 μg/m3 is thus proposed for IMs. Evaluation of an additional PoD - benchmark dose lower confidence limit (BMDL) - further supported the default HBELs. The default oral PDE can also serve as a threshold of toxicological concern (TTC) for IMs. Default limits can aid in setting HBELs for novel data-poor IMs, as well as supporting waiving of RDT in the future through read-across.

A Novel Machine Learning Model and a Web Portal for Predicting the Human Skin Sensitization Effects of Chemical Agents

Skin sensitization is a significant concern for chemical safety assessments. Traditional animal assays often fail to predict human responses accurately, and ethical constraints limit the collection of human data, necessitating a need for reliable in silico models of skin sensitization prediction. This study introduces HuSSPred, an in silico tool based on the Human Predictive Patch Test (HPPT). HuSSPred aims to enhance the reliability of predicting human skin sensitization effects for chemical agents to support their regulatory assessment. We have curated an extensive HPPT database and performed chemical space analysis and grouping. Binary and multiclass QSAR models were developed with Bayesian hyperparameter optimization. Model performance was evaluated via five-fold cross-validation. We performed model validation with reference data from the Defined Approaches for Skin Sensitization (DASS) app. HuSSPred models demonstrated strong predictive performance with CCR ranging from 55 to 88%, sensitivity between 48 and 89%, and specificity between 37 and 92%. The positive predictive value (PPV) ranged from 84 to 97%, versus negative predictive value (NPV) from 22 to 65%, and coverage was between 75 and 93%. Our models exhibited comparable or improved performance compared to existing tools, and the external validation showed the high accuracy and sensitivity of the developed models. HuSSPred provides a reliable, open-access, and ethical alternative to traditional testing for skin sensitization. Its high accuracy and reasonable coverage make it a valuable resource for regulatory assessments, aligning with the 3Rs principles. The publicly accessible HuSSPred web tool offers a user-friendly interface for predicting skin sensitization based on chemical structure.

Evaluating the consistency of judgments derived through both in silico and expert application of the Cramer classification scheme

The Cramer classification scheme has emerged as one of the most extensively-adopted predictive toxicology tools, owing in part to its employment for chemical categorisation within threshold of toxicological concern evaluation. The characteristics of several of its rules have contributed to inconsistencies with respect to degree of hazard attributed to common (particularly food-relevant) substances. This investigation examines these discrepancies, and their origins, raising awareness of such issues amongst users seeking to apply and/or adapt the rule-set. A dataset of over 3000 compounds was assembled, each with Cramer class assignments issued by up to four groups of industry and academic experts. These were complemented by corresponding outputs from in silico implementations of the scheme present within Toxtree and OECD QSAR Toolbox software, including a working of a "Revised Cramer Decision Tree". Consistency between judgments was assessed, revealing that although the extent of inter-expert agreement was very high (≥97%), general concordance between expert and in silico calls was more modest (∼70%). In particular, 22 chemical groupings were identified to serve as prominent sources of disagreement, the origins of which could be attributed either to differences in subjective interpretation, to software coding anomalies, or to reforms introduced by authors of the revised rules.