Low-cost quantum mechanical descriptors for data efficient skin sensitization QSAR models.

Area of Science:

• Computational chemistry
• Toxicology
• Quantitative Structure Activity Relationship (QSAR) modelling

Background:

• QSAR models require mechanistic data for accuracy, particularly electrophilic reactivity for skin sensitization.
• High computational costs of traditional quantum mechanics limit dataset size.
• Low-cost ab initio methods are needed for accurate QSAR modelling of skin sensitization.

Purpose of the Study:

• To investigate the use of low-cost Hf-3c electronic descriptors for QSAR modelling of skin sensitization.
• To assess the performance of Hf-3c descriptors in predicting in vitro and in vivo skin sensitization assay outcomes.
• To model the Ames assay as a surrogate endpoint for skin sensitization.

Main Methods:

• Calculated electronic descriptors using the Hartree Fock with 3 corrections (Hf-3c) method.
• Employed the conductor-like polarizable continuum model (CPCM) for implicit solvation.
• Developed QSAR models for in vitro Ames, KeratinoSens, and Direct Peptide Reactivity Assay (DPRA) datasets.
• Validated models using in vivo Local Lymph Node Assay (LLNA) and Human Repeated Insult Patch Test (HRIPT) data.

Main Results:

• Hf-3c descriptors with CPCM improved QSAR model performance for Ames (AUC=0.770), KeratinoSens (AUC=0.763), and DPRA (AUC=0.750) assays.
• Combined models achieved high predictive performance for unseen LLNA (AUC=0.789) and HRIPT (AUC=0.791) data.
• Hf-3c offers higher chemical accuracy than previous semiempirical methods at lower computational cost.

Conclusions:

• Low-cost Hf-3c quantum mechanical descriptors enhance QSAR model validity and predictive performance for skin sensitization.
• This approach enables the development of robust QSAR models using larger datasets.
• The Hf-3c method provides a cost-effective alternative for mechanistic QSAR modelling in toxicology.