Predicting fluorescence to singlet oxygen generation quantum yield ratio for BODIPY dyes using QSPR and machine learning

Functional dyes that are capable of both bright fluorescence and efficient singlet oxygen generation are crucial for theranostic techniques, which integrate fluorescence imaging and photodynamic therapy (PDT). The development of new functional dyes for theranostics is often costly and time-consuming due to laborious synthesis and post-synthetic screening of large libraries of compounds. In this work, we describe machine learning methods suitable for simultaneous prediction of fluorescence and photosensitizing ability of heavy-atom-free boron dipyrromethene (BODIPY) compounds. We analysed the ratio between fluorescence quantum yield (ΦFl) and singlet oxygen quantum yield (ΦΔ) for over 70 BODIPY structures in polar (acetonitrile) and non-polar (toluene) solvents, which mimic hydrophilic and hydrophobic cell environments, respectively. QSPR models were developed based on more than 5000 calculated molecular descriptors, including quantum chemical and topological descriptors. We applied multiple linear regression (MLR), support vector regression (SVR), and random forest regression (RFR) methods for model building and optimization. The resulting models demonstrated robust statistical parameters (R2 = 0. 73-0. 91) for both polar and non-polar media. The relative contributions of the descriptors to the models were assessed, identifying Eig03EA (dm), F01C-N, and TDB06p as the most influential. These results demonstrate that QSPR machine learning methods are effective in predicting key photochemical parameters of BODIPY photosensitizers, thereby potentially streamlining the development of theranostic agents.