Abstract

Biosynthetic gene clusters (BGCs) encode enzymatic pathways for natural products with pharmaceutical potential, yet prioritizing candidates from fragmented environmental DNA (eDNA) assemblies remains computationally challenging. We present BGC-QDR (Biosynthetic Gene Cluster Quantum Discovery and Ranking), an open-source pipeline that integrates input quality control, Prodigal ORF prediction, Pfam HMM domain annotation, rule-based BGC classification, MiBIG 4.0 novelty assessment, and variational quantum classifier (VQC) ranking via PennyLane. BGC-QDR is designed as a quantum-assisted ranking framework for biologically informed BGC prioritization, not as a claim of quantum computational advantage over classical machine learning. We evaluate the pipeline on MiBIG 4.0 (2,636 annotated BGCs) using a 20-dimensional biosynthetic feature vector and stratified 10-fold cross-validation. The integrated VQC (6 qubits x 3 layers, 54 parameters) achieves an accuracy of 0.789 +/- 0.076 and ROC-AUC of 0.835 +/- 0.057. Random Forest achieves the highest ROC-AUC (0.898 +/- 0.032), followed by Logistic Regression (0.874 +/- 0.020) and MLP (0.872 +/- 0.024). Wilcoxon signed-rank tests on per-fold AUC scores show that VQC ROC-AUC is significantly lower than Random Forest (p = 0.0098) and Logistic Regression (p = 0.037) at alpha = 0.05, with no significant difference versus MLP (p = 0.064). Architecture ablation identifies 4 qubits x 3 layers as the best VQC configuration on hold-out validation (AUC = 0.737). Feature importance analysis highlights peptidyl carrier protein domains, cluster length, and module count as dominant predictors. BGC-QDR provides a reproducible, end-to-end workflow for eDNA-derived BGC discovery with integrated novelty scoring and quantum-assisted candidate ranking.