Quantum Simulation of Preferred Tautomeric State Prediction

Yu Shee, Tzu-Lan Yeh, Jen-Yueh Hsiao, Ann Yang, Yen‐Chu Lin, Min-Hsiu Hsieh

Abstract

Prediction of tautomers plays an essential role in computer-aided drug discovery. However, it remains a challenging task nowadays to accurately predict the canonical tautomeric form of a given drug-like molecule. Lack of extensive tautomer databases, most likely due to the difficulty in experimental studies, hampers the development of effective empirical methods for tautomer predictions. A more accurate estimation of the stable tautomeric form can be achieved by quantum chemistry calculations. Yet, the computational cost required prevents quantum chemistry calculation as a standard tool for tautomer prediction in computer-aided drug discovery. In this paper we propose a hybrid quantum chemistry-quantum computation workflow to efficiently predict the dominant tautomeric form. Specifically, we select active-space molecular orbitals based on quantum chemistry methods. Then we utilize efficient encoding methods to map the Hamiltonian onto quantum devices to reduce the qubit resources and circuit depth. Finally, variational quantum eigensolver (VQE) algorithms are employed for ground state estimation where hardware-efficient ansatz circuits are used. To demonstrate the applicability of our methodology, we perform experiments on two tautomeric systems: acetone and Edaravone, each having 52 and 150 spin-orbitals in the STO-3G basis set, respectively. Our numerical results show that their tautomeric state prediction agrees with the CCSD benchmarks. Moreover, the required quantum resources are efficient: in the example of Edaravone, we could achieve chemical accuracy with only eight qubits and 80 two-qubit gates.

Introduction

In drug discovery, the biological activity and solubility of a molecule are heavily dependent on its specific isomeric state. Many drug-like molecules exist as tautomers—isomers that interconvert through proton migration and double-bond rearrangement. Identifying the most stable (preferred) tautomeric form is a cornerstone of computer-aided drug design (CADD).

Current empirical methods often struggle due to a lack of experimental data, while high-level classical quantum chemistry methods like Coupled Cluster theory (CCSD) are computationally prohibitive for large molecules. Quantum computing offers a potential solution, but current Noisy Intermediate-Scale Quantum (NISQ) hardware is limited by low qubit counts and high error rates. This research addresses these limitations by developing a resource-efficient hybrid workflow.

Methodology

The proposed methodology utilizes a six-step hybrid quantum-classical approach designed to minimize the quantum resources required for accurate electronic structure calculations.

The hybrid quantum-classical workflow for tautomer prediction.

The workflow begins with classical Geometry Optimization using Density Functional Theory (DFT). To reduce the problem size, Active Space Selection is performed to focus on molecular orbitals (MOs) most relevant to the tautomerization process. A critical innovation is the use of Qubit-Efficient Encoding (QEE), which maps the Hamiltonian onto qubits logarithmically rather than linearly, significantly reducing the hardware requirements compared to standard methods like Jordan-Wigner (JW) encoding. The Variational Quantum Eigensolver (VQE) then estimates the ground state energy using hardware-efficient ansatz circuits and optimized initialization strategies, such as Gaussian initialization, to avoid convergence issues like barren plateaus.

Results and Discussion

The methodology was validated on two systems: acetone (a simple keto-enol system) and Edaravone (an FDA-approved drug for ALS with keto, enol, and amine forms).

Comparison of quantum resources required for different molecules and encoding schemes.

As shown in the resource table, the reduction in qubit requirements is substantial. For Edaravone, which possesses 150 spin-orbitals in the STO-3G basis, standard JW encoding would require 150 qubits. By utilizing an active space of 12 spin-orbitals and applying QEE, the requirement was reduced to just 8 qubits.

Numerical simulations demonstrated that the quantum workflow correctly identified the keto form as the most stable for both molecules, matching classical CCSD benchmarks. For Edaravone, chemical accuracy was achieved using only 80 two-qubit (CNOT) gates. Even under simulated hardware noise (T1/T2 relaxation), the shallow-circuit approach remained robust, maintaining results near the chemical accuracy threshold (~1 kcal/mol).

Conclusion

This work demonstrates a viable path for applying NISQ-era quantum computers to practical drug discovery tasks. By combining active-space selection with qubit-efficient encoding, complex drug molecules like Edaravone can be simulated with minimal quantum resources. Future improvements will focus on incorporating solvent effects via Continuum Solvation Models and automating the active space selection process to enhance scalability for larger, more flexible drug candidates.

Summary updated

Abstract

Prediction of tautomers plays an essential role in computer-aided drug discovery. However, it remains a challenging task nowadays to accurately predict the canonical tautomeric form of a given drug-like molecule. Lack of extensive tautomer databases, most likely due to the difficulty in experimental studies, hampers the development of effective empirical methods for tautomer predictions. A more accurate estimation of the stable tautomeric form can be achieved by quantum chemistry calculations. Yet, the computational cost required prevents quantum chemistry calculation as a standard tool for tautomer prediction in computer-aided drug discovery. In this paper we propose a hybrid quantum chemistry-quantum computation workflow to efficiently predict the dominant tautomeric form. Specifically, we select active-space molecular orbitals based on quantum chemistry methods. Then we utilize efficient encoding methods to map the Hamiltonian onto quantum devices to reduce the qubit resources and circuit depth. Finally, variational quantum eigensolver (VQE) algorithms are employed for ground state estimation where hardware-efficient ansatz circuits are used. To demonstrate the applicability of our methodology, we perform experiments on two tautomeric systems: acetone and Edaravone, each having 52 and 150 spin-orbitals in the STO-3G basis set, respectively. Our numerical results show that their tautomeric state prediction agrees with the CCSD benchmarks. Moreover, the required quantum resources are efficient: in the example of Edaravone, we could achieve chemical accuracy with only eight qubits and 80 two-qubit gates.
BibTeX
@inproceedings{shee2023quantum,
  author = {Yu Shee and Tzu-Lan Yeh and Jen-Yueh Hsiao and Ann Yang and Yen‐Chu Lin and Min-Hsiu Hsieh},
  title = {{Quantum Simulation of Preferred Tautomeric State Prediction}},
  booktitle = {Npj},
  year = {2023},
  doi = {10.1038/s41534-023-00767-9},
}
Figure