Quantum Use Case #2: Protein folding and binding-affinity estimation

Short answer: This use case involves applying hybrid quantum-classical workflows to estimate the binding affinity between proteins and ligands, a critical step in drug discovery and repositioning. While researchers have demonstrated that current Noisy Intermediate Scale Quantum (NISQ) devices can rank-order inhibitors based on calculated binding energy differences, the approach remains at the research stage. It is not yet an operational tool for commercial drug design, but it offers a promising pathway for tackling computationally challenging problems involving strongly correlated electronic systems.

The Computational Challenge in Binding Affinity

Determining how tightly a drug candidate binds to a target protein is a central problem in medicinal chemistry. Traditional computational methods often struggle with precise binding affinity prediction, particularly when the target involves complex electronic structures. For instance, amyloid beta, an intrinsically disordered protein implicated in neurodegenerative diseases like Alzheimer’s, exhibits binding affinities to metal centers such as iron and copper. These interactions are governed by strong electron correlation, making the calculation of ground state energies a significant bottleneck for classical numerical methods.

The difficulty lies in accurately estimating the ground state energy of these molecular systems. Classical algorithms face exponential scaling issues as the system size grows, limiting their ability to model strongly correlated centers with high fidelity. This computational barrier creates an opportunity for quantum computing, which theoretically offers advantages in simulating quantum mechanical systems. However, realizing this advantage requires understanding the specific quantum resources needed to outperform classical baselines. Current research focuses on identifying these resource requirements to determine whether quantum speedup is feasible for such binding affinity calculations.

Hybrid Quantum-Classical Workflows

To address these challenges, researchers have proposed and demonstrated hybrid workflows that combine classical embedding techniques with quantum variational algorithms. One prominent approach utilizes the Density Matrix Embedding Theory (DMET) procedure alongside the Variational Quantum Eigensolver (VQE). DMET allows for the embedding of molecular fragments into a larger system, while VQE is used to find the molecular electronic ground states. This combination creates a prototypical workflow for quantifying protein-ligand interactions.

In recent studies, this hybrid approach has been applied to a series of beta-secretase (BACE1) inhibitors. By calculating binding energy differences, researchers were able to rank-order these inhibitors. The authors describe this work as the first application of real quantum computers to the calculation of protein-ligand binding energies. The experiments utilized the latest NISQ devices, including superconducting transmon and trapped-ion platforms. These demonstrations show that it is technically possible to execute such workflows on current hardware, although the results are primarily intended to shed light on the capabilities and limitations of the technology rather than to provide immediate pharmaceutical solutions.

Current Hardware and Research-Stage Limitations

The research indicates that these calculations are performed on NISQ devices, which are characterized by limited qubit counts and higher error rates compared to future fault-tolerant machines. The use of superconducting transmon and trapped-ion hardware highlights the diversity of quantum platforms being explored for this application. However, the fact that these are NISQ devices implies that the calculations are subject to noise and require careful error mitigation or verification strategies.

It is crucial to frame these results as research-stage achievements. While the rank-ordering of BACE1 inhibitors is a significant milestone, it does not imply that quantum computers are currently superior to classical methods for routine drug discovery tasks. The primary goal of such studies is to validate the quantum workflow and understand the quantum resources required. For example, in the case of amyloid beta, the focus is on understanding the resource requirements for calculating binding affinities to metal centers, rather than on delivering a clinically viable therapeutic intervention. The complexity of the problem, involving strongly correlated metal centers, makes it a suitable candidate for quantum acceleration, but the practical utility remains under investigation.

Implications for Drug Repositioning

Beyond de novo drug design, binding affinity estimation is a key strategy in drug repositioning. This approach identifies new uses for approved or investigational drugs that are outside their original medical indication. Drug repositioning is generally less expensive and time-consuming than traditional drug discovery. A critical step in this process is determining the binding affinity of existing drugs to new target proteins.

Deep learning methods have recently demonstrated superior performance in precise binding affinity prediction compared to conventional computational methods, driven by the large increase in available experimental data. However, these classical deep learning approaches can be complex and time-consuming. Quantum machine learning approaches, such as hybrid quantum neural networks, are being explored as alternatives that may offer efficiency gains or improved accuracy for specific subsets of problems. While classical deep learning currently dominates this space, the integration of quantum components into neural network architectures for binding affinity prediction remains an active area of research. The potential benefit lies in leveraging quantum effects to handle the high-dimensional feature spaces involved in molecular interactions, though this remains largely theoretical or in early simulation stages.

FAQ

Is quantum computing currently used for commercial drug discovery?

No. The available research indicates that these applications are at the research stage. While hybrid workflows have been demonstrated on NISQ devices, they are not yet deployed as standard tools in commercial pharmaceutical development. The primary objective is to validate the technology and understand its limits.

What specific quantum hardware has been used for protein-ligand calculations?

Researchers have utilized the latest NISQ devices, specifically superconducting transmon and trapped-ion platforms. These devices were used to calculate binding energy differences for a series of BACE1 inhibitors, which the authors describe as the first application of real quantum computers to this specific type of calculation.

How does the hybrid workflow function?

The workflow combines the Density Matrix Embedding Theory (DMET) with the Variational Quantum Eigensolver (VQE). DMET handles the embedding of molecular fragments, while VQE is employed to find the molecular electronic ground states. This combination allows for the quantification of protein-ligand interactions by estimating the relevant energy differences.

Why is amyloid beta a challenging target for classical computation?

Amyloid beta involves binding affinities to metal centers like iron and copper, which are characterized by strong electron correlation. Numerically calculating these affinities is computationally challenging because obtaining accurate estimates of the ground state energy for such strongly correlated systems is a bottleneck for classical algorithms.

Does quantum computing replace deep learning in binding affinity prediction?

Not necessarily. Deep learning methods currently demonstrate superior performance in precise binding affinity prediction due to large experimental datasets. Quantum approaches, such as hybrid quantum neural networks, are being explored as complementary or alternative methods, but they have not yet displaced classical deep learning in practical applications.