PSoSOQY: A Deep Learning-Driven Singlet Oxygen Quantum Yield Prediction Platform for Expediting Photosensitizer...

The singlet oxygen quantum yield (SOQY) of photosensitizers (PSs) is a critical parameter in photodynamic therapy (PDT), directly linked to therapeutic efficacy by quantifying singlet oxygen generation. Traditional approaches to developing high-SOQY PSs are complex, involving elaborate molecular design and challenging SOQY measurements, which hinder rapid therapeutic advancement. This study constructs a comprehensive dataset to develop BiLSTM + attention (BA)-SOQY, a cutting-edge prediction model integrating bidirectional long short-term memory (BiLSTM) networks and attention mechanisms. BiLSTM captures long-term dependencies in sequential data, while the attention mechanism highlights critical input features, enabling efficient screening of potential PSs candidates. On the held-out test set, BA-SOQY achieves R² = 0.9140; validations on ESOL and FreeSolv datasets yield R² > 0.9, supporting robustness and generalization. Additionally, substructure masking interpretation strategy based on SMILES (SMIS-SMILES) is introduced. Combined with BA-SOQY, this forms the PSoSOQY platform, which not only provides accurate SOQY predictions but also reveals how substructures influence SOQY. These insights accelerate rational PSs design and high-throughput screening, driving progress in PDT research.