Fragment-based prediction of skin sensitization using recursive partitioning

Area of Science:

• Toxicology
• Computational Chemistry
• Chemical Risk Assessment

Background:

• Skin sensitization is a critical toxicological endpoint for chemical risk assessment.
• Predicting skin sensitization potential is crucial for ensuring chemical safety and guiding drug design.

Purpose of the Study:

• To develop and validate a structure-activity relationship (SAR) model for predicting chemical skin sensitization.
• To identify key structural fragments and physicochemical properties associated with skin sensitization.

Main Methods:

• Utilized GASTON (GrAph/Sequence/Tree extractiON) to extract structural fragments from 357 compounds with Local Lymph Node Assay (LLNA) data.
• Constructed a recursive partitioning (RP) tree model using selected fragment descriptors and physicochemical properties.
• Validated the model using a leave-one-out approach on training and independent test sets.

Main Results:

• Identified eight significant structural fragments contributing to skin sensitization prediction.
• Achieved high balanced accuracy: 0.846 (training set), 0.800 (test set I), and 0.809 (test set II).
• The fragment-based RP tree demonstrated superior performance in identifying skin sensitizers.

Conclusions:

• Fragment-based RP tree analysis is a robust method for predicting skin sensitization.
• The identified fragments offer insights into sensitization mechanisms and guide the design of less sensitizing chemicals.
• This approach provides valuable guidance for chemical risk assessment and the development of safer chemical products.