QUANTUM-ENHANCED PATHOGENICITY PREDICTION OF HBB GENE VARIANTS: A HYBRID QUANTUM-CLASSICAL MACHINE LEARNING FRAMEWORK FOR Β-THALASSEMIA

Abstract

Beta-thalassemia is a prevalent monogenic haemoglobin disorder caused by pathogenic variants in the HBB gene. Prior work established a classical deep learning framework achieving ROC-AUC 0.9483 using nine biologically informed features. Classical models remain constrained in modelling high-dimensional nonlinear feature interactions inherent in genomic data. This study extends the classical framework with three quantum machine learning components: a Quantum Kernel Support Vector Machine (QKSVM), a Hybrid Quantum-Classical Neural Network (HQNN), and a Quantum Explainability module integrating SHAP GradientExplainer with quantum input-gradient attribution. A ClinVar-derived dataset of 1,585 HBB single nucleotide variants (1,323 benign, 262 pathogenic) was used throughout. The QKSVM employs a depth-2 ZZFeatureMap fidelity kernel on four qubits encoding top SHAP-identified features (REVEL, AlphaMissense, PolyPhen-2, CADD). The HQNN retains a classical residual encoder and replaces the output layer with a parameterised four-qubit variational circuit trained via adjoint differentiation. Spearman rank correlation measured concordance between explainability methods. The QKSVM achieved ROC-AUC 0.9332 and PR-AUC 0.6364, substantially outperforming the classical RBF-SVM baseline (ROC-AUC 0.8706, PR-AUC 0.4836). The HQNN achieved ROC-AUC 0.9210 and PR-AUC 0.6599 with pathogenic recall of 0.96. Quantum explainability identified REVEL as the dominant predictor across both methods, while SIFT rose markedly in quantum gradient attribution. Spearman correlation between methods was ρ = 0.517, indicating moderate agreement. To our knowledge, this is the first application of quantum machine learning to HBB-specific variant pathogenicity prediction. The QKSVM surpasses its classical SVM counterpart, consistent with theoretical expectations for NISQ-era performance on small tabular genomic datasets. The quantum explainability framework offers novel multi-method attribution with direct implications for clinical variant interpretation.