Quantum-Enhanced Neural Networks for Rare Disease Diagnosis: A Comparative Evaluation Against Classical Machine Learning

This paper develops and benchmarks a hybrid quantum-enhanced neural network (QENN) against classical machine learning models for rare disease variant classification. Combining angle-based quantum encoding, a parameterized quantum circuit, and classical post-processing, the QENN was evaluated against five classical/deep learning baselines and a matched classical neural network across ClinVar-derived variant data. XGBoost achieved the strongest classical performance (accuracy 0.9575, ROC-AUC 0.9728), while QNN variants matched but did not exceed a comparable classical neural network (accuracy/F1 ≈ 0.88, ROC-AUC up to 0.95), at substantially higher computational cost. These results provide an empirically grounded answer to whether quantum-enhanced architectures offer an advantage for rare disease diagnosis, showing that a clear quantum benefit remains unestablished at current feature scales.