How to use quantum computers for biomolecular free energies

Jakob Günther, Thomas Weymuth, Moritz Bensberg, Freek Witteveen, Matthew S. Teynor, F. Emil Thomasen, Valentina Sora, William Bro‐Jørgensen, Raphael T. Husistein, Mihael Eraković, Marek Miller, Leah P. Weisburn, Minsik Cho, Marco Eckhoff, Aram W. Harrow, Anders Krogh, Troy Van Voorhis, Kresten Lindorff‐Larsen, Gemma C. Solomon, Markus Reiher, Matthias ChristandlarXiv 2025 · DOI · arXiv

TL;DR

This paper introduces FreeQuantum, a novel computational pipeline that bridges the gap between highly accurate quantum mechanical calculations and large-scale biomolecular simulations. By combining a multi-layer quantum embedding strategy with machine learning, FreeQuantum can efficiently leverage quantum computing to predict crucial biomolecular free energies, paving the way for more accurate drug discovery and biological understanding.

What Problem Does This Paper Address?

Predicting how molecules interact – a fundamental aspect of biology and drug design – relies heavily on calculating "free energies." These calculations quantify the energy changes involved in processes like a drug binding to a protein. While accurate quantum mechanical (QM) methods can provide precise energy values, they are computationally prohibitive for the large molecules involved in biological systems. Traditional molecular mechanics (MM) force fields, while scalable, lack the accuracy needed for many critical interactions, especially those involving transition metals or complex electronic structures. This creates a significant challenge: how to achieve high accuracy for large biomolecules when the most accurate methods are limited to tiny systems?

Methods

The core of this work is the FreeQuantum pipeline, which employs a three-layer QM/QM/MM embedding strategy:

  1. MM Layer: Classical molecular mechanics force fields are used for the bulk of the biomolecular system, enabling efficient sampling of configurations.
  2. QM/MM Layer: A "quantum region" is defined, encompassing key interaction sites (e.g., the drug-protein interface). Here, QM calculations (initially DFT) are performed on selected snapshots.
  3. QM/QM/MM Layer (Quantum Cores): Within the QM region, even smaller "quantum cores" are identified. These cores are treated with the highest level of accuracy, using traditional high-level QM methods (like coupled cluster or NEVPT2) or, in the future, quantum computers.

To bridge the gap between these layers and handle the scarcity of high-accuracy QM data, machine learning (ML) is crucial:

  • ML1 Potential: An ML potential is trained using the QM/MM data to approximate the QM energies and forces across the entire quantum region. This is done within an active learning loop to iteratively improve the ML model by focusing on regions of high uncertainty.
  • ML2 Potential: The ML1 potential is further refined using the highly accurate QM/QM/MM data from the quantum cores. This transfer learning approach allows the ML model to learn from the most accurate data points and improve its predictions for the entire quantum region.

The pipeline is designed to be automated and modular, allowing for easy integration of different QM engines and embedding strategies.

Datasets

The paper uses data from:

  • A ruthenium-based anticancer drug (NKP-1339) binding to a protein (GRP78/NKP-1339): This is the primary benchmark system for demonstrating the FreeQuantum pipeline.
  • A previously studied system (MCL-1/19G): This system, involving an organic closed-shell ligand, was used for validation and comparison with experimental and coupled cluster data.

The paper does not introduce a new benchmark dataset but rather applies its methodology to a challenging, relevant biological system.

Key Insights and Ideas

  • Quantum Embedding is Key: The multi-layer QM/QM/MM embedding is essential for isolating the most critical electronic interactions while treating the rest of the system classically. This allows for a manageable computational cost.
  • Machine Learning as a Bridge: ML potentials are not just interpolators; they are crucial for transferring high-accuracy QM information from small quantum cores to the larger QM region and ultimately to the entire biomolecular system. This is especially important when high-accuracy QM data is scarce.
  • Quantum Computing's Role: The paper explicitly analyzes the quantum resources (qubits, gate depth, error rates) required for quantum computers to provide the necessary high-accuracy energy calculations for the quantum cores. This provides a roadmap for achieving "quantum advantage" in this field.
  • Modularity and Flexibility: The FreeQuantum pipeline is designed to be highly modular, allowing for the easy swapping of QM engines (traditional or quantum) and embedding strategies, making it adaptable to future advancements.

Key Results

  • The FreeQuantum pipeline successfully calculated the binding free energy of a ruthenium-based anticancer drug to its protein target.
  • The pipeline demonstrated the ability to refine free energy calculations from a pure MM description (-19.1 ± 1.5 kJ/mol) to QM/MM with DFT (-17.0 ± 2.6 kJ/mol) and finally to QM/QM/MM with high-level wavefunction methods (-11.3 ± 2.9 kJ/mol).
  • The high-level QM/QM/MM results showed a significant deviation (6-8 kJ/mol) from DFT and MM results, highlighting the importance of accurate electronic structure treatment for such systems.
  • The paper provides concrete estimates for the quantum resources (qubits, gate counts, error tolerance) needed for quantum computers to achieve chemical accuracy for the quantum cores in this system, suggesting that around 1000 logical qubits with low error rates would be required for a 20-minute computation per data point.

Limitations and Future Work

  • Current QM Limitations: While the pipeline is designed for future quantum computers, current traditional QM methods still have limitations in terms of accuracy and scalability for very large quantum cores.
  • Active Learning for ML2: The paper notes that while ML1 training benefits from forces, ML2 training currently relies solely on energies. Incorporating forces into ML2 training could further improve accuracy.
  • Automated QM Region Selection: While an automated algorithm for QM region selection is implemented, the current pipeline run used manual selection. Future work could leverage this automation.
  • Quantum Hardware Development: The realization of quantum advantage is contingent on the development of fault-tolerant quantum computers with sufficient qubits and low error rates.
  • Generalization: While the pipeline is designed to be general, further validation on a wider range of biomolecular systems, especially those with strong electronic correlations, would be beneficial.
  • Future Work: The authors emphasize the need for continued development of quantum algorithms and hardware to fully realize the potential of quantum computing for biomolecular free energy calculations. They also highlight the open-source nature of FreeQuantum, encouraging community contributions.

Summary updated

Abstract

Free energy calculations are at the heart of physics-based analyses of biochemical processes. They allow us to quantify molecular recognition mechanisms, which determine a wide range of biological phenomena from how cells send and receive signals to how pharmaceutical compounds can be used to treat diseases. Quantitative and predictive free energy calculations require computational models that accurately capture both the varied and intricate electronic interactions between molecules as well as the entropic contributions from motions of these molecules and their aqueous environment. However, accurate quantum-mechanical energies and forces can only be obtained for small atomistic models, not for large biomacromolecules. Here, we demonstrate how to consistently link accurate quantum-mechanical data obtained for substructures to the overall potential energy of biomolecular complexes by machine learning in an integrated algorithm. We do so using a two-fold quantum embedding strategy where the innermost quantum cores are treated at a very high level of accuracy. We demonstrate the viability of this approach for the molecular recognition of a ruthenium-based anticancer drug by its protein target, applying traditional quantum chemical methods. As such methods scale unfavorable with system size, we analyze requirements for quantum computers to provide highly accurate energies that impact the resulting free energies. Once the requirements are met, our computational pipeline FreeQuantum is able to make efficient use of the quantum computed energies, thereby enabling quantum computing enhanced modeling of biochemical processes. This approach combines the exponential speedups of quantum computers for simulating interacting electrons with modern classical simulation techniques that incorporate machine learning to model large molecules.
BibTeX
@misc{gunther2025how,
  author = {Jakob Günther and Thomas Weymuth and Moritz Bensberg and Freek Witteveen and Matthew S. Teynor and F. Emil Thomasen and Valentina Sora and William Bro‐Jørgensen and Raphael T. Husistein and Mihael Eraković and Marek Miller and Leah P. Weisburn and Minsik Cho and Marco Eckhoff and Aram W. Harrow and Anders Krogh and Troy Van Voorhis and Kresten Lindorff‐Larsen and Gemma C. Solomon and Markus Reiher and Matthias Christandl},
  title = {{How to use quantum computers for biomolecular free energies}},
  year = {2025},
  eprint = {2506.20587},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2506.20587},
  howpublished = {arXiv},
}
Figure