Predicting DNA Reactions with a Quantum Chemistry-Based Deep Learning Model

DNA reactions are crucial in biology, synthetic biology, and DNA computing. Accurate prediction of thermodynamic and kinetic parameters is vital for understanding molecular interactions and designing functional DNA-based systems. Existing models have limitations due to simplifications and approximations that may deviate from experimental measurements. In this study, we propose a quantum chemistry-based deep learning model to enhance accuracy and efficiency in predicting DNA reaction parameters. The model integrates quantum chemistry calculations, new designed descriptor matrices, and deep learning algorithms. It comprehensively describes energy variations by expanding stacks and considering relevant factors. To address limited labeled data, an active learning method selects informative samples iteratively, optimizing data utilization. The results demonstrate the superior predictive capabilities of our model in accurately determining DNA hybridization free energies and strand displacement rate constants. This integration of quantum chemistry and deep learning improves our understanding of DNA reactions and facilitates precise design and optimization of DNA-based systems.