Quantum Heuristics Optimise Protein Sequence Design on Noisy Intermediate-Scale Devices

Designing new proteins with desired properties represents a major challenge in biotechnology, and researchers are now exploring the potential of quantum computing to accelerate this process. Hanna Linn, Lucas Knuthson, and Anders Irbäck, from Chalmers University of Technology and Lund University, alongside Sandipan Mohanty et al. from Forschungszentrum Jülich, investigate how variational quantum algorithms can tackle the initial step of protein design, identifying amino acid sequences that favour a specific three-dimensional structure. Their work demonstrates that, while early quantum approaches struggle with the inherent noise of current quantum computers, circuits designed to better match hardware limitations offer improved performance in both simulations and on real devices. This represents a significant step towards harnessing the power of quantum computing to create novel proteins with applications ranging from medicine to materials science, and highlights the importance of tailoring algorithms to the realities of near-term quantum technology.

Quantum heuristics demonstrate promise in solving diverse optimisation problems, including lattice protein folding. Equally relevant is the inverse problem, protein design, where researchers seek sequences that fold to a given target structure. This work investigates the utility of variational quantum algorithms for the first of these two steps on today’s noisy intermediate-scale quantum devices, focusing on the sequence optimisation task which is less resource-demanding than folding computations.

Protein Folding with Variational Quantum Algorithms

This document details research into the application of Variational Quantum Algorithms (VQAs), specifically Quantum Approximate Optimization Algorithm (QAOA) and related techniques, to the problem of protein folding and design. Researchers are exploring whether near-term quantum computers can offer advantages in tackling this computationally challenging problem, crucial for drug discovery and materials science. The work includes analysis of the limitations and potential of current quantum hardware for these applications. The core problem is to find the lowest energy conformation of a protein given its amino acid sequence, or to design a sequence that folds into a desired structure.

This is computationally expensive due to the vast number of possible protein shapes, so researchers use simplified protein models to make the problem tractable for current quantum computers. The team explores QAOA, a VQA designed for complex optimisation problems, transforming the protein folding/design problem into a quadratic unconstrained binary optimisation (QUBO) problem. They investigate different quantum circuit architectures, varying the number of layers and types of quantum gates used, and employ classical optimisation algorithms to find the best settings for the quantum circuit. Addressing challenges like barren plateaus is crucial, as is the representation of the protein’s structure or sequence as qubits, explored through different encoding schemes.

Initial work is done using quantum simulators to test algorithms and parameters, followed by experiments on real quantum computers to assess performance on actual hardware, with techniques to mitigate noise and improve accuracy. A key challenge is scaling the algorithms to handle larger proteins, explored through reducing the number of qubits required. The research utilises Qiskit, IBM’s open-source quantum computing framework, along with Python, NumPy, and SciPy for implementation and analysis. If successful, these algorithms could accelerate drug discovery by enabling the design of proteins with specific properties, and contribute to the development of new materials. Future work will explore more advanced quantum algorithms, efficient encoding schemes, and error correction techniques.

Variational Algorithms Optimize Simplified Protein Folding

Researchers are exploring the potential of variational quantum algorithms (VQAs) to tackle complex optimisation problems, specifically focusing on protein design, the task of creating amino acid sequences that fold into desired structures. This work investigates whether these algorithms can efficiently find the lowest energy sequences for a given protein structure, a crucial step in the design process, concentrating on a simplified model of protein folding to reduce computational demands while retaining key challenges. The study compares two approaches to building quantum circuits for this optimisation task. One method, based on QAOA variants, incorporates knowledge of the protein structure directly into the circuit design, but requires significant depth, making it impractical for current quantum hardware.

The other approach, utilising a Hardware Efficient Ansatz (HEA), prioritises compatibility with the limitations of real quantum devices by creating circuits tailored to the available gate set and connectivity, proving more robust even in the presence of simulated noise. Interestingly, the team discovered that transferring optimised parameters from similar protein design problems, known as parameter donation, significantly improved the performance of the HEA approach. However, when tested on actual quantum hardware, the IBM Torino device, this scheme failed to deliver satisfactory results, highlighting that current noise models used in simulations do not fully capture the complex and time-dependent nature of noise present in real quantum computers. The team attributes this failure to unsimulated noise features, suggesting that more sophisticated noise models are needed to accurately predict algorithm performance and guide the development of robust quantum solutions.

HEA Excels in Noisy Protein Optimization

This research investigates the application of variational quantum algorithms to the problem of protein sequence optimisation, a crucial step in protein design. The team tested two approaches, QAOA and HEA, aiming to identify sequences that minimise energy for a given protein structure. Results demonstrate that while QAOA performs well in ideal simulations, its performance significantly degrades with noise, limiting its practical application on current quantum hardware. In contrast, HEA, designed with simpler circuits, exhibits greater resilience to noise, achieving improved performance in both simulated and real quantum environments.

The study successfully applied HEA to optimise short protein chains (up to length 12) on a real quantum device, representing a step towards using quantum computers for protein design. However, discrepancies between simulated and real device performance suggest that the simulated noise models do not fully capture the complexities of hardware noise, particularly its temporal aspects. Future work should focus on developing more accurate noise models and exploring error mitigation techniques to unlock the full potential of these algorithms, especially QAOA, for solving more complex protein design challenges. The research also suggests that parameter reuse strategies could improve the efficiency of these algorithms by leveraging knowledge gained from solving related problems.