Predicting the quantum yield of ¹O₂ generation for pteridines and fluoroquinolones using machine learning

Area of Science:

• Photochemistry and Photophysics
• Computational Chemistry
• Medicinal Chemistry

Background:

• Fluoroquinolones (FQs) and Pterins (Ptrs) are aza-bicyclic compounds with significant photochemical activity, including singlet oxygen generation.
• Their similar electronic absorption spectra suggest potential for unified photochemistry studies.
• Understanding photosensitizing ability is crucial for applications like photodynamic therapy.

Purpose of the Study:

• To develop and compare machine learning (ML) methods for predicting the photosensitizing ability (singlet oxygen generation quantum yield, ΦΔ) of FQs and Ptrs.
• To identify key molecular descriptors influencing photosensitization in these diverse compounds.
• To assess the feasibility of unified modeling for structurally different photosensitizers.

Main Methods:

• Investigated singlet oxygen generation quantum yield (ΦΔ) for 48 pterins and fluoroquinolones in deuterated water.
• Utilized a dataset of over 5000 molecular descriptors, reduced via a genetic algorithm (GA).
• Employed and compared multiple linear regression (MLR), support vector regression (SVR), random forest regression (RFR), gradient boosting (GBR), and extreme gradient boosting (XGBoost) for ML model development.

Main Results:

• Achieved high predictive performance across models (Rtrain² > 0.97, q² > 0.84).
• Support vector regression (SVR) showed the best test set performance (Rtest² = 0.975), while extreme gradient boosting (XGBoost) demonstrated superior robustness.
• Interpretability analyses (SHAP, ALE) identified conjugated maximum bond length (CMBL) and long-range polarizability (TDB09p) as critical descriptors for ΦΔ.

Conclusions:

• Demonstrated the feasibility of unified machine learning modeling for structurally diverse photosensitizers like FQs and Ptrs.
• Provided mechanistic insights into photosensitization through descriptor relevance analysis.
• Offered actionable insights for the rational design and virtual screening of novel photosensitizers for photodynamic therapy.