Hybrid quantum neural networks with variational quantum regressor for enhancing QSPR modeling of CO2-capturing amine

Abstract

Accurate amine property prediction is essential for optimizing CO₂ capture efficiency in post-combustion processes. Quantum machine learning (QML) can enhance predictive modeling by leveraging superposition, entanglement, and interference to capture complex correlations. In this study, we developed hybrid quantum neural networks (HQNN) to improve quantitative structure-property relationship (QSPR) modeling for CO₂-capturing amines. By integrating variational quantum regressors with classical multi-layer perceptrons and graph neural networks, quantum-enhanced performance was explored in physicochemical property prediction under noiseless conditions and robustness was evaluated against quantum hardware noise using IBM quantum systems. Our results showed that HQNNs improve predictive accuracy for key solvent properties, including basicity, viscosity, boiling point, melting point, and vapor pressure. The fine-tuned and frozen pre-trained HQNN models with 9 qubits consistently achieved the highest rankings, highlighting the benefits of integrating quantum layers with pre-trained classical models. Furthermore, simulations under hardware noise confirmed the robustness of HQNNs, maintaining predictive performance. Overall, these findings emphasize the potential of hybrid quantum-classical architectures in molecular modeling. As quantum hardware and QML algorithms continue to advance, practical quantum benefits in QSPR modeling and materials discovery are expected to become increasingly attainable, driven by improvements in quantum circuit design, noise mitigation, and scalable architectures.