A Hybrid Quantum-Classical Workflow for Early-Stage Molecular Discovery: A Workflow-Level Proof-of-Concept

Authors

  • Charnelle Razo Department of Informatics and Analytics, National University of Science and Technology, Zimbabwe Author
  • Belinda Ndlovu Department of Informatics and Analytics, National University of Science and Technology, Zimbabwe Author https://orcid.org/0000-0001-6046-3240

DOI:

https://doi.org/10.15294/sji.v13i3.50262

Keywords:

Hybrid Quantum-classical Systems, Molecular Screening, VQE Optimisation, Quantum Feature Encoding, Workflow integration, MolecularInteractionFeatureMap (MIFM)

Abstract

Purpose: Classical computational approaches struggle to model molecular quantum interactions, making drug discovery costly and time-consuming. Quantum computing shows promise in molecular modelling, but much of the existing work focuses on isolated proof-of-concept tasks. These studies usually do not examine the full molecular discovery process. This study demonstrates the feasibility of a unified hybrid quantum–classical workflow for early-stage molecular discovery and examines how quantum-based molecular representations influence later optimisation and prediction.

Methods/Study design/approach: The study developed and evaluated a five-stage hybrid quantum-classical workflow integrating query interpretation, molecular screening with quantum re-ranking, VQE-based energy estimation,  property prediction and report generation. The system was implemented using Qiskit, Qiskit Nature, RDKit, and spaCy on the QM9 dataset using noiseless state-vector simulation.

Result/Findings: The NLP component was evaluated across four tiers: 180 template-based queries, 160 independently written queries, 1,000 independently written queries, and 1,000 externally sourced BioASQ Task B questions. F1-scores were 89.9%, 82.1%, 73.0%, and 71.3%. Screening achieved constraint satisfaction rates of 90%-100%. Quantum-informed ranking differed substantially from cosine similarity and evaluated the utility of this through top-k enrichment and multi-property hit rates. MIFM achieved a mean VQE absolute error of 0.0156 Ha compared with 0.0218 Ha for ZZFeatureMap and 0.0248 Ha for cosine similarity. VQE and property-prediction evaluations, scalability experiments, and simulated NISQ tests provided broader evidence of workflow behaviour.

Novelty/Originality/Value: This study presents a reproducible hybrid quantum–classical workflow for early-stage molecular discovery, extended with a chemistry-aware MolecularInteractionFeatureMap (MIFM). The study does not claim universal quantum advantage or physical quantum-hardware validation.

Downloads

Published

29-08-2026

Article ID

50262

Issue

Section

Articles

How to Cite

A Hybrid Quantum-Classical Workflow for Early-Stage Molecular Discovery: A Workflow-Level Proof-of-Concept. (2026). Scientific Journal of Informatics, 13(3), 727-742. https://doi.org/10.15294/sji.v13i3.50262