HIV-1 protease cleavage sites detection with a quantum convolutional neural network algorithm.

Human immunodeficiency virus type 1 (HIV-1) protease plays a crucial role in viral maturation by cleaving both viral and host precursor proteins. Accurate prediction of HIV-1 protease cleavage sites is therefore essential for understanding viral pathogenesis and developing therapeutic inhibitors. This study aimed to establish a quantum convolutional neural network (QCNN)-based framework integrated with a neural quantum embedding (NQE) to predict HIV-1 protease cleavage sites from amino acid sequences of viral and human proteins. The proposed framework combines QCNN and NQE to enhance feature representation in the quantum space. We evaluated the model using four publicly available HIV-1 protease cleavage site datasets. To examine the performance and robustness of the quantum model, we compared it with classical neural networks under both noiseless and noisy quantum simulation environments. The experiments were conducted using different numbers of qubits and trainable parameter scales to assess scalability and parameter efficiency. Across all experimental conditions, QCNN models incorporating NQE with angle and amplitude encoding achieved higher classification accuracy than classical neural networks. The average accuracy values of the 4-qubit and 8-qubit QCNNs were 0.9146 and 0.8929, respectively, outperforming the classical neural networks with average accuracies of 0.6125 and 0.8278. Moreover, the QCNN integrated with NQE using the ZZ feature map and angle encoding maintained relatively stable classification performance under the simulated quantum hardware noise conditions evaluated in this study. This study presents the first application of an NQE-augmented QCNN framework for HIV-1 protease cleavage site prediction. The results demonstrate that certain quantum neural architectures can outperform parameter-matched classical counterparts under the simulated noise conditions evaluated in this study. The findings suggest that NQE-enhanced QCNNs hold strong potential for scalable, noise-resilient quantum machine learning in biomedical sequence classification and could provide a foundation for future quantum-based bioinformatics analyses.