Global structure-activity relationship model for nonmutagenic carcinogens using virtual ligand-protein interactions...

Area of Science:

• Toxicology and Cheminformatics
• Computational toxicology
• Structure-Activity Relationship (SAR) modeling

Background:

• Structure-activity relationship (SAR) models are crucial for understanding chemical carcinogen mechanisms and predicting carcinogenicity.
• Existing SAR models often struggle with nonmutagenic carcinogens, highlighting a need for improved predictive approaches.

Purpose of the Study:

• To compare a traditional fragment-based SAR approach with a novel virtual ligand-protein interaction-based approach for modeling nonmutagenic carcinogens.
• To evaluate the performance of these models using distinct learning sets, including data from the Carcinogenic Potency Database and rodent cancer bioassays.

Main Methods:

• Developed ligand-based SAR models utilizing computationally calculated ligand-binding affinities to 5495 proteins.
• Created two learning sets: one from the Carcinogenic Potency Database and another from Malacarne et al. for nonalerting compounds.
• Assessed model performance based on mutagenicity categorization and concordance rates for nonmutagenic/nonalerting carcinogens and noncarcinogens.

Main Results:

• The traditional fragment model outperformed the ligand-based model when rat cancer models were categorized by mutagenicity.
• For nonmutagenic or nonalerting compounds, the fragment model showed ~50% concordance, while ligand-based models achieved 71% (nonmutagenic) and 74% (nonalerting).

Conclusions:

• Virtual ligand-protein interaction analysis shows potential for developing predictive SAR models for nonmutagenic carcinogens.
• A hybrid approach, using fragment-based models for mutagenic chemicals and ligand-based models for nonmutagenic ones, may offer a practical strategy for carcinogenesis SAR modeling.