Relationship between molecular connectivity and carcinogenic activity: a confirmation with a new software program...

Area of Science:

• Computational toxicology
• Cheminformatics
• Structure-Activity Relationships (SAR)

Background:

• Carcinogenicity assessment is crucial for public health.
• Predictive toxicology models can accelerate chemical safety evaluations.
• Identifying structural alerts for carcinogenicity is a key challenge.

Purpose of the Study:

• To develop and validate a novel computational method for predicting chemical carcinogenicity.
• To explore the utility of contiguous-atom fragments in structure-activity relationship (SAR) modeling.
• To assess the predictive accuracy of fragment-based carcinogenicity prediction.

Main Methods:

• A database of 826 chemicals tested for carcinogenicity was used.
• Chemical structures were fragmented into contiguous-atom fragments (2-8 non-hydrogen atoms) using graph theory software.
• An 80/20 training/test split was employed, with 8 independent runs for statistical robustness.
• Fragments significantly associated with carcinogenicity (p < 0.125) were identified from training sets.

Main Results:

• An average of 315 significant fragments were identified per training set.
• The model predicted carcinogenicity for 77% of test set molecules.
• Average prediction accuracy was 67.5% for the test set.
• Chemicals with only positive fragments achieved 78.7% accuracy, while those with mixed or negative fragments were around 60% accurate.
• Randomly assigned carcinogenicity in pseudo-training sets yielded no predictive power in test sets.

Conclusions:

• The developed fragment-based SAR approach shows significant potential for predicting chemical carcinogenicity.
• The method provides a valid and reproducible way to identify structural features associated with toxicological endpoints.
• This approach can complement traditional methods for chemical safety assessment and drug discovery.