Predictive Modeling of DNA Damage Outcomes: Classification of Mutational Determinants Using Augmented Machine...

Area of Science:

• Computational toxicology
• Molecular biology
• Machine learning in bioinformatics

Background:

• DNA damage from chemical exposure can lead to mutations, influenced by damage structure, replication, and repair mechanisms.
• Factors like adduct size, flanking bases, and DNA polymerase type are critical in determining mutational outcomes.
• Existing in vitro and in vivo studies highlight the complexity of predicting these outcomes.

Purpose of the Study:

• To develop machine learning (ML) models for predicting the mutational outcome of DNA adducts encountered by polymerases.
• To identify key parameters influencing DNA mutational outcomes.
• To enhance prediction accuracy using generative network methods and advanced classifiers.

Main Methods:

• Trained Logistic Regression (LR), Decision Tree (DT), and Support Vector Machine (SVM) models using literature data on DNA adducts and abasic sites.
• Utilized a generative network (Bootstrapped-Variational Autoencoder - BT-VAE) to create synthetic data mirroring real-world DNA damage scenarios.
• Employed an Extreme Gradient Boosting Classifier to determine the most influential parameters affecting mutational outcomes.

Main Results:

• The proposed BT-VAE model significantly enhanced classifier prediction accuracy by 40%.
• The Support Vector Machine (SVM) model demonstrated superior performance across all evaluated classification metrics.
• The BT-VAE framework successfully predicted mutational outcomes (match/mismatch for adducts, adenine/nonadenine for abasic sites) using adduct size, polymerase, and flanking base as inputs.

Conclusions:

• Machine learning, particularly the BT-VAE framework combined with SVM, offers a powerful tool for predicting DNA mutational outcomes.
• This approach provides valuable insights for in vivo and in vitro toxicology studies.
• Accurate prediction of mutation types aids in understanding the mechanisms of chemical mutagenesis and carcinogenicity.