Drug design on quantum computers

The paper addresses how and where future quantum computers could impact drug discovery, focusing on quantum chemistry computations as the most likely first practical application. The context is the escalating cost and timeline of drug development and the increasing role of computation in guiding design. Classical simulations—from machine learning and molecular dynamics to quantum mechanical methods—either lack sufficient accuracy or are computationally prohibitive for many relevant tasks. While quantum computers are expected to enable efficient simulation of quantum systems, current algorithmic efforts largely target strongly correlated electronic structure problems where classical methods fail. The authors argue that if quantum advantage is confined to such systems, its practical significance for drug design may be limited, given most drug-like molecules are weakly correlated. The perspective aims to evaluate the current status, map quantum algorithm capabilities to drug design needs, and outline research directions required to make quantum computing broadly impactful in pharma.

Status of quantum computing for chemistry: Present-day hardware is in the NISQ regime, where algorithms like VQE rely on classical optimization and suffer from unfavorable measurement scaling, making runtime estimation difficult. Consequently, the analysis focuses on fault-tolerant quantum computers (FTQCs) with quantum error correction, which impose substantial space-time overheads (e.g., order of millions of physical qubits for classically challenging targets). Quantum advantage is expected for strongly correlated electronic structure problems; identifying such systems requires demanding diagnostics (multi-reference character, spin-symmetry breaking, natural orbital occupations, entanglement growth, and failures of cluster expansions). Ground-state energies can be obtained with state preparation plus quantum phase estimation (QPE), with cost strongly dependent on initial state overlap. Prior resource estimates (e.g., for FeMoco) have improved from years to days through algorithmic advances, indicating a trajectory toward practical FTQC chemistry. The literature also covers methods for forces and observables, highlighting that further algorithmic, error-correction, and hardware progress is needed for industrial relevance.

As a perspective, no empirical experiments are conducted. The paper outlines a canonical FTQC workflow for electronic structure: (1) classical preprocessing to refine molecular geometry, construct and optimize the Hamiltonian representation, and design an error-corrected circuit; (2) quantum initial-state preparation of a classically selected approximate ground state; (3) QPE to project onto eigenstates and extract the ground-state energy, with runtime scaling inversely with the initial overlap. The workflow can be extended to compute other observables (e.g., forces) and properties. The article details practical constraints: error-correction overhead (thousands of physical qubits per logical qubit), the centrality of state preparation quality for QPE runtime, and the need for compact Hamiltonian representations and improved error-correction schemes. For drug design tasks (e.g., binding free energies), the paper reviews classical methodologies (force fields, DFT, CC, alchemical perturbations, and ensemble sampling via molecular dynamics) and frames how quantum algorithms might accelerate or augment them, particularly moving beyond single-point energies toward thermodynamic properties derived from ensembles or time evolution.

• FTQCs, not NISQ devices, are required for quantum chemistry calculations that could impact industrial drug design due to noise and measurement-scaling limitations on NISQ.
• Quantum advantage is most evident for strongly correlated electronic structure problems, identifiable via multi-reference character, spin-symmetry issues, natural orbital occupations, entanglement scaling, and known failure points of cluster expansions. However, many drug-like molecules are weakly correlated, limiting immediate applicability if focus remains only on strong correlation.
• QPE-based workflows can, in principle, compute ground-state energies without uncontrolled approximations, but cost depends critically on initial state overlap.
• Resource estimates illustrate current scale: a challenging target such as FeMoco requires roughly 200 logical qubits mapped to about 2 million physical qubits; algorithmic improvements have reduced projected runtimes from years to days for such targets.
• Drug design’s key computational need—accurate binding free energies—requires ensemble sampling and potentially billions of single-point energy/force evaluations. Classical force fields are fast but often unreliable; DFT/CC are more accurate but too costly for routine free-energy workflows.
• At physiological temperatures, a 1.5 kcal/mol error in binding free energy can lead to approximately an order-of-magnitude error in dose predictions, underscoring the need for high accuracy (≈1.0 kcal/mol to experiment).
• Quantum computing could have modest impact on ancillary tasks (reaction mechanism optimization, NMR/IR/VCD spectra) relative to the potential value of faster, accurate design-stage energetics.
• Major bottlenecks include error-correction overhead, initial state preparation, Hamiltonian representation size, and the lack of heuristics with analyzable scaling for weakly correlated, large systems.
• The greatest impact will require methods going beyond single-point energies to compute thermodynamic properties efficiently, potentially trading accuracy for speed or reducing sampling demands.

The analysis aligns quantum computing capabilities with the actual needs of drug discovery. While QC promises polynomial-time solutions for strongly correlated systems, most industrially relevant drug-like molecules are weakly correlated and embedded in large thermodynamic ensembles (protein-ligand-solvent systems with thousands of atoms). Thus, a singular focus on exact ground-state energies of strongly correlated fragments underutilizes QC’s potential for drug design. The paper argues that to meaningfully influence lead optimization, quantum algorithms must help with the computation of free energies and related thermodynamic properties at costs competitive with or better than classical approximations, while providing improved robustness and accuracy. Progress in error correction, initial state preparation strategies, compact Hamiltonians, and possibly heuristic quantum algorithms tailored to weak correlation and large systems would broaden applicability. By enabling accurate, faster predictions, QC could replace cycles of labor-intensive experiments and make in silico design more predictive, but only if algorithmic and hardware advances extend beyond niche strongly correlated cases.

Current classical computing methods fail to describe quantum systems accurately enough in relevant times for the pharmaceutical industry, limiting the applicability of quantum chemistry to drug design. More accurate computations could bring significant value to the pharmaceutical industry by replacing many labour-intensive experiments with calculations in silico, as long as the computational cost is lower than the experimental effort. Quantum computations could enable key, experimentally inaccessible insights into chemical systems, exploiting methods that directly derive properties from wave functions. To have a profound impact on the pharmaceutical industry, quantum computers need to benefit a broader set of problems than the small number inaccessible to classical computers. Typical relevant systems have thousands of atoms and rarely require exact accuracy, but many pharmaceutical use cases rely on thermodynamic ensembles requiring many single-point calculations. New methods that trade accuracy for time on quantum computers or avoid sampling could be beneficial. Ideally, quantum computers should offer accuracy and robustness for both strongly and weakly correlated systems at speeds currently only accessible by lower-accuracy methods. Continued algorithmic advances (state preparation, Hamiltonian representations), hardware, and error-correction improvements are necessary to move beyond single-point energies. Open research between academia and industry will be crucial to make quantum computing an essential tool for faster, better drug design.

• Perspective article: no empirical benchmarking or experimental validation; conclusions synthesize existing literature and projections.
• Applicability depends on future availability of large-scale fault-tolerant quantum computers; resource estimates are subject to change with evolving hardware and algorithms.
• Identification of strongly correlated systems is nuanced and relies on expert-driven diagnostics; generality across diverse drug targets may vary.
• Heuristic quantum methods for weakly correlated, large systems are not yet theoretically characterized or practically benchmarked due to the absence of error-corrected hardware.
• Free energy and ensemble properties on quantum hardware remain conceptual; concrete algorithms with proven advantage and end-to-end costings are not established.