Reliable CA-(Q)SAR generation based on entropy weight optimized by grid search and correction factors

Area of Science:

• Computational chemistry and toxicology
• Artificial intelligence in drug discovery
• Genotoxicity assessment

Background:

• Chromosome aberration (CA) is a significant genotoxicity linked to carcinogenicity and developmental issues.
• Accurate prediction of CA is crucial for drug safety evaluation.
• Existing methods require improvement for reliable genotoxicity screening.

Purpose of the Study:

• To develop a robust Quantitative Structure-Activity Relationship (QSAR) model for predicting chromosome aberration (CA) genotoxicity.
• To leverage artificial intelligence (AI) and diverse molecular descriptors for enhanced prediction accuracy.
• To identify structural features associated with CA genotoxicity for informed drug design.

Main Methods:

• Construction of a QSAR model using machine learning and deep learning algorithms on a dataset of 3208 compounds.
• Optimization of algorithms through hyperparametric iterations and integration of molecular fingerprints and drug-like properties (MP) using entropy weight methodology.
• Inclusion of molecular similarity and molecular connection index for improved prediction of similar compounds.
• Development on an open-source Python platform.

Main Results:

• The final CA-(Q)SAR model demonstrated a prediction accuracy of 80.6% with a bias of approximately 0.9793.
• Analysis identified typical structural features within numerical intervals (MPI) of molecular properties (MW, XlogP, TPSA) associated with CA toxicity.
• Normalized occurrence probability (NOP) analysis highlighted key structural determinants for CA genotoxicity.

Conclusions:

• The developed AI-driven QSAR model provides a reliable tool for predicting CA genotoxicity.
• The findings offer valuable insights into structural alerts for CA genotoxicity, guiding the design of safer drug candidates.
• This approach can help in the early screening of potential drug leads to avoid genotoxic liabilities.