Source-linked AI summary
Drug design on quantum computers
Raffaele Santagati, Alan Aspuru-Guzik, Ryan Babbush, Matthias Degroote, Leticia Gonzalez, Elica Kyoseva, Nikolaj Moll, Markus Oppel, Robert M. Parrish, Nicholas C. Rubin, Michael Streif, Christofer S. Tautermann, Horst Weiss, Nathan Wiebe, Clemens Utschig-Utschig
TL;DR
Drug development faces rising costs, while computational methods for chemical systems can be limited by speed and accuracy. This perspective examines where quantum chemistry could contribute to drug design and the challenges to practical use. It concludes that quantum computations could provide chemical insights, with industrial value conditional on computational cost.
Problem
Chemical simulations, including quantum-mechanical effects, remain computationally intensive and can be limited in practical applicability by speed and accuracy.
Source: INTRODUCTION · p. 1Method
The perspective assesses quantum-chemistry problems where quantum computers could offer speed-ups, compares them with computer-aided drug-design needs, and discusses research directions.
Source: INTRODUCTION · p. 1Results
The perspective identifies quantum chemistry as a potential early quantum-computing application for drug design, while noting that practical quantum advantage requires major hardware and algorithm progress.
Source: INTRODUCTION · p. 1Takeaways & Limitations
More accurate in-silico computations could replace some labour-intensive experiments if their computational cost is lower than experimental effort.
Source: CONCLUSION · p. 6The lack of error-corrected quantum computers prevents thorough benchmarking of heuristic algorithms today.
Source: §III · p. 5
Abstract
from arXiv · showhide
Quantum computers promise to impact industrial applications, for which quantum chemical calculations are required, by virtue of their high accuracy. This perspective explores the challenges and opportunities of applying quantum computers to drug design, discusses where they could transform industrial research and elaborates on what is needed to reach this goal.
INTRODUCTION
Drug development costs have risen sharply, while computational methods for simulating chemical systems face speed and accuracy limits. This perspective examines where quantum chemistry could help drug design and what would be required to realize that potential.
INTRODUCTION
- Drug-development costs rose from tens of millions of dollars in the 1950s to billions today, even after inflation adjustment [1].Computational approaches are increasingly used in research and development, but chemical simulations remain computationally intensive.Source: INTRODUCTION · p. 1
- Current computational methods for chemical systems, including quantum-mechanical effects, can be limited in practical applicability by speed and accuracy.Methods used in research and development include machine learning, molecular dynamics, and quantum-mechanical calculations.Source: INTRODUCTION · p. 1
- Quantum computers have been proposed to simulate quantum systems efficiently, motivating investment in quantum chemistry applications [9–19].The research community has developed quantum hardware and algorithms, while most efforts target challenging electronic-structure problems.Source: INTRODUCTION · p. 1
- Strongly correlated systems may offer quantum advantage, but identifying them is difficult and their limited scope could constrain drug-design relevance.Indicators exist, but verifying them can require substantial time and chemical expertise.Source: INTRODUCTION · p. 1
- The perspective focuses on quantum-chemistry calculations as potential early quantum-computing applications capable of affecting drug design.It considers where quantum chemistry might speed up calculations relative to classical methods, compares those problems with computer-aided drug-design needs, and discusses research directions for pharmaceutical use.Source: INTRODUCTION · p. 1
I. STATUS QUO: QUANTUM COMPUTERS
This section outlines fault-tolerant quantum computing for electronic-structure calculations, including resource estimation, ground-state energy workflows, and the conditions relevant to potential drug-design applications.
I. STATUS QUO: QUANTUM COMPUTERS
- Computational-cost estimates specify required qubits and run-time, providing engineering targets for hardware and identifying algorithmic needs.The section focuses on fault-tolerant quantum computers because NISQ algorithms rely heavily on classical optimization heuristics, have difficult-to-estimate run-times, and may require exponentially more measurements as circuit depth increases.Source: §I · pp. 1–2
- Five indicators distinguish cases that might require quantum computing, including multi-reference character, essential spin-symmetry breaking, cluster-expansion failure, near-degenerate orbitals, and extensive orbital entanglement.The indicators include wavefunctions needing many comparable-amplitude determinants and non-integer natural-orbital occupations.Source: §I · p. 2
- Quantum computers are expected to help solve electronic-structure problems in strongly correlated systems, but verifying that a system meets the relevant conditions can be demanding.Such systems include multi-metal systems with similar metal electronic environments and interactions.Source: §I · p. 2
- The workflow uses classical preprocessing to refine geometry, the Hamiltonian, and an initial state before generating circuits and applying QPE to estimate ground-state energy.The prepared state is a superposition with high ground-state overlap, and measurement projects it into the ground state.Source: §I · p. 2, Figure 1 · p. 3
- Ground-state energy estimation becomes progressively more expensive as the initial state's overlap with the correct ground state decreases.The calculation combines state preparation with quantum phase estimation.Source: §I · p. 2
- Algorithmic improvements reduced FeMoco ground-state energy run-time estimates from years to days, although FTQC algorithms cannot yet be executed.A classically challenging FeMoco simulation was estimated to require roughly 200 logical qubits implemented in 2 million physical qubits, beyond current hardware.Source: §I · p. 2
II. COMPUTER-AIDED DRUG DESIGN
Drug discovery narrows candidates through years of testing and refinement, while computational chemistry supports pharmacokinetic prediction and binding-affinity calculations. The central challenge is obtaining accurate binding and thermodynamic predictions across large molecular systems at practical computational cost.
II. COMPUTER-AIDED DRUG DESIGN
- Drug discovery starts with a target and narrows weak binders through years of optimization toward efficacious, safe oral candidates.Thousands of molecules may be synthesized in hit-to-lead and lead-optimization programs before candidates are identified for clinical development.Source: §II · p. 3
- Computational chemistry supports pharmacokinetic-property prediction and binding-affinity calculations, with affinity translating into projected therapeutic dose.Alchemical perturbation methods estimate relative binding strength by gradually morphing a known compound into a new one.Source: §II · pp. 3–4
- Binding calculations must capture many accessible geometries and binding pathways because the minimum-free-energy configuration is the statistically most frequent one.Sampling these ensembles entails many single-point calculations, and explicit solvent adds degrees of freedom and complexity.Source: §II · p. 4
- Higher-accuracy electronic-structure methods become applicable to smaller systems, from mean-field approaches through coupled-cluster methods to finite-basis exact full configuration interaction.Force fields and semi-empirical methods can be tuned using quantum-method information; coupled-cluster expands around a single mean-field reference.Source: §II · p. 4
- Full DFT treatment of binding free energies remains elusive at current costs, while coupled-cluster calculations are out of scope beyond small systems.A binding free-energy calculation can take many hours classically, and higher accuracy can increase costs by several orders of magnitude.Source: §II · p. 5
- Quantum-chemistry speedups are most relevant to lead optimization, while most oral drugs are small closed-shell molecules that generally lack strong correlation.The perspective describes other quantum use cases, including reaction-mechanism optimization and molecular spectra, as having more modest potential impact; lower-accuracy DFT often predicts NMR spectra well.Source: §II · p. 5
III. CHALLENGES AND PROSPECTS
The perspective argues that quantum computing’s drug-design impact depends on progress beyond single-point calculations for a small set of strongly correlated systems. It identifies hardware and algorithmic bottlenecks, alongside possible routes toward useful calculations of thermodynamic properties and broader applications.
III. CHALLENGES AND PROSPECTS
- Quantum computers may speed electronic-structure calculations for strongly correlated systems, but broader drug-design impact requires going beyond single-point energies.The perspective cites better understanding of cytochrome P450 as one possible application.Source: §III · p. 5
- Quantum error correction is a dominant fault-tolerant runtime and space overhead; FeMoco ground-state calculations are estimated to require millions of physical qubits.Reducing overhead calls for lower hardware error rates, increased qubit connectivity, and improvements to error correction.Source: §III · p. 5
- Preparing the initial state remains an unresolved challenge because QPE runtime depends on it and on its overlap with the target state.Heuristic solutions have been proposed, and subsystem decomposition is a potential approach for weakly correlated systems.Source: §III · p. 5
- More compact Hamiltonian representations could reduce runtime, but the lack of error-corrected computers prevents thorough benchmarking of heuristic algorithms.The perspective also proposes analyzing heuristic scaling and constant factors for specific input parameters.Source: §III · p. 5
- For industrial applications, faster approximate methods may matter more in the mid-term than highest-accuracy calculations, but outperforming linear-scaling DFT and Hartree-Fock is difficult.Possible directions include using quantum results to design better DFT functionals, accelerating tensor-network contractions, or implementing coupled-cluster optimization.Source: §III · pp. 5–6
- Billions of single-point calculations, each potentially taking days on a quantum computer, make current thermodynamic-property calculations impractical.One proposed route is to model classical nuclei and electrons together and generate thermal geometry ensembles to calculate free energies directly.Source: §III · p. 6
- Quantum machine learning on wave functions from large quantum computers is a speculative route to predicting pharmacokinetic properties.The perspective also notes that quantum computers could compute wave functions for ensembles of molecules.Source: §III · p. 6
CONCLUSION
Quantum computing could make drug-design calculations more accurate and predictive, but broader impact depends on handling large systems and thermodynamic ensembles while achieving practical computational costs. Hardware, error-correction, and algorithmic advances remain necessary to move beyond single-point energies.
CONCLUSION
- More accurate quantum calculations could replace labour-intensive experiments when their computational cost is lower, and reveal experimentally inaccessible chemical insights.Current classical methods do not describe relevant quantum systems accurately enough within timescales relevant to the pharmaceutical industry.Source: CONCLUSION · p. 6
- Drug-design calculations ideally need accuracy and robustness for strongly and weakly correlated systems, at speeds currently available only from lower-accuracy methods.Relevant systems can have thousands of atoms, and thermodynamic properties may require many single-point calculations across large ensembles.Source: CONCLUSION · p. 6
- Further hardware, error-correction, and algorithm advances are needed for practical industry applications and to extend quantum calculations beyond single-point energies.Electronic-structure algorithm advances have already reduced computational costs.Source: CONCLUSION · p. 6
- Open research integrating academia and industry is presented as a route toward using quantum computing to design better drugs faster.The perspective notes that steps are already being made and several routes exist to address the challenges.Source: CONCLUSION · p. 6