[Paper Review] Quantum Machine Learning Framework for Virtual Screening in Drug Discovery: a Prospective Quantum Advantage

Stefano Mensa, Emre Sahin+3 more|arXiv (Cornell University)|Apr 8, 2022

Quantum Computing Algorithms and Architecture56 references4 citations

TL;DR

This paper proposes a hybrid classical-quantum machine learning framework for ligand-based virtual screening in drug discovery, combining classical Support Vector Classifiers with quantum kernel estimation. It demonstrates prospective quantum advantage by showing that quantum-enhanced classification outperforms classical methods on real-world ADRB2 and COVID-19 datasets using both simulations and IBM Quantum hardware, with AUC-ROC results matching statevector simulations and surpassing classical baselines.

ABSTRACT

Machine Learning (ML) for Ligand Based Virtual Screening (LB-VS) is an important in-silico tool for discovering new drugs in a faster and cost-effective manner, especially for emerging diseases such as COVID-19. In this paper, we propose a general-purpose framework combining a classical Support Vector Classifier (SVC) algorithm with quantum kernel estimation for LB-VS on real-world databases, and we argue in favor of its prospective quantum advantage. Indeed, we heuristically prove that our quantum integrated workflow can, at least in some relevant instances, provide a tangible advantage compared to state-of-art classical algorithms operating on the same datasets, showing strong dependence on target and features selection method. Finally, we test our algorithm on IBM Quantum processors using ADRB2 and COVID-19 datasets, showing that hardware simulations provide results in line with the predicted performances and can surpass classical equivalents.

Motivation & Objective

• To develop a general-purpose quantum machine learning framework for ligand-based virtual screening (LB-VS) in drug discovery.
• To investigate whether quantum kernel methods can provide a tangible advantage over classical machine learning algorithms on real-world drug discovery datasets.
• To validate the performance of quantum-enhanced classifiers on near-term quantum hardware using real molecular datasets.
• To establish the concept of Prospective Quantum Advantage (PQA) as a stepping stone toward full quantum advantage in large-scale drug discovery workflows.

Proposed method

• The framework uses RDKit to extract cheminformatics features from SMILES-encoded molecules in drug databases.
• Feature selection and reduction techniques (ANOVA, PCA) are applied to reduce dimensionality and improve classification performance.
• A classical Support Vector Classifier (SVC) is trained using both classical and quantum kernels derived from quantum circuits.
• Quantum kernel estimation is performed via quantum circuits implemented on IBM Quantum processors (Montreal and Guadalupe) and simulated on quantum hardware emulators.
• The performance of the quantum-enhanced classifier (QSVC) is evaluated using AUC-ROC as the primary metric and compared to classical baselines, including RMD and MPNN models.
• Numerical simulations using statevector methods validate the expected performance before hardware execution, ensuring reliable prediction of quantum hardware outcomes.

Experimental results

Research questions

• RQ1 Can a quantum kernel-based classifier outperform classical machine learning models in ligand-based virtual screening tasks?
• RQ2 Does the proposed hybrid quantum-classical framework demonstrate a measurable advantage on real-world drug discovery datasets when executed on near-term quantum hardware?
• RQ3 To what extent do quantum simulations accurately predict the performance of quantum algorithms on actual quantum processors?
• RQ4 Can the framework be scaled to larger feature sets that remain classically intractable, thereby enabling prospective quantum advantage?

Key findings

• The QSVC model trained with a quantum kernel achieved an AUC-ROC score on the ADRB2 dataset that was comparable to statevector simulations and fell within their error bars when executed on IBM Quantum Montreal.
• For the COVID-19 dataset, the QSVC model executed on IBM Quantum Guadalupe produced an AUC-ROC result that matched the best-performing quantum numerical simulations.
• Quantum hardware results for both ADRB2 and COVID-19 significantly outperformed all classical baseline methods, including RMD and MPNN models.
• Simulations on quantum hardware emulators accurately predicted the outcomes of actual quantum processor runs, validating the reliability of numerical simulations for future quantum advantage assessment.
• The framework demonstrates Prospective Quantum Advantage (PQA), indicating that observed quantum benefits in small-scale instances are likely to scale to larger, classically infeasible problems.
• The study confirms that quantum kernel methods are a viable and superior approach for classifying active and inactive molecules in drug discovery, especially when feature selection is optimized.