Quantum Algorithm for Protein Structure Prediction Using the Face-Centered Cubic Lattice

Rui-Hao Li ,, Hakan Doga ,, Bryan Raubenolt ,, Sarah Mostame, Nicholas DiSanto, Fabio Cumbo, Jayadev Joshi, Hanna Linn, Maeve Gaffney, Alexander Holden, Vinooth Kulkarni, Vipin Chaudhary, Kenneth M. Merz Jr, Abdullah Ash Saki, Tomas Radivoyevitch, Frank DiFilippo, Jun Qin, Omar Shehab, Daniel Blanke · 2025

Read Quantum Algorithm for Protein Structure Prediction Using the Face-Centered Cubic Lattice by Rui-Hao Li ,, Hakan Doga ,, Bryan Raubenolt ,, Sarah Mostame, Nicholas DiSanto, Fabio Cumbo, Jayadev Joshi, Hanna Linn, Maeve Gaffney, Alexander Holden, Vinooth Kulkarni, Vipin Chaudhary, Kenneth M. Merz Jr, Abdullah Ash Saki, Tomas Radivoyevitch, Frank DiFilippo, Jun Qin, Omar Shehab, Daniel Blanke (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-08955.txt.

Introduction

Predicting how a protein folds is a search so large that quantum computers are being tried on toy versions. A large team runs the first quantum folding algorithm on a face-centered cubic lattice, a grid roomy enough to mimic real protein backbones better than simpler lattices. Two solvers, a polynomial fit method and a constrained variational one, attack a six-link chain.

Both ran on real IBM hardware, an older Eagle chip and a newer Heron. Despite noise, they recovered the chain's lowest-energy shapes. The newer machine sampled the right answer up to three times more often. The 2025 experiment measures hardware progress through a biology problem.

Catalog record

CREATORRui-Hao Li ,, Hakan Doga ,, Bryan Raubenolt ,, Sarah Mostame, Nicholas DiSanto, Fabio Cumbo, Jayadev Joshi, Hanna Linn, Maeve Gaffney, Alexander Holden, Vinooth Kulkarni, Vipin Chaudhary, Kenneth M. Merz Jr, Abdullah Ash Saki, Tomas Radivoyevitch, Frank DiFilippo, Jun Qin, Omar Shehab, Daniel Blanke
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE92 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-08955

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Quantum Algorithm for Protein Structure Prediction Using the Face-Centered Cubic Lattice Rui-Hao Li ,, Hakan Doga ,, Bryan Raubenolt ,, Sarah Mostame, Nicholas DiSanto, Fabio Cumbo, Jayadev Joshi, Hanna Linn, Maeve Gaffney, Alexander Holden, Vinooth Kulkarni, Vipin Chaudhary, Kenneth M. Merz Jr, Abdullah Ash Saki, Tomas Radivoyevitch, Frank DiFilippo, Jun Qin, Omar Shehab, Daniel Blankenberg , Abstract In this work, we present the first implementation of the face-centered cubic (FCC) lattice model for protein structure prediction with a quantum algorithm. Our motivation to encode the FCC lattice stems from our observation that the FCC lattice is more capable in terms of modeling realistic secondary structures in proteins compared to other lattices, as demonstrated using root mean square deviation (RMSD). We utilize two quantum methods to solve this problem: a polynomial fitting approach (PolyFit) and the Variational Quantum Eigensolver with constraints (VQEC) based on the Lagrangian duality principle. Both methods are successfully deployed on Eagle R3 ( ibm_cleveland ) and Heron R2 ( ibm_kingston ) quantum computers, where we are able to recover ground state configurations for the 6-amino acid sequence KLVFFA under noise. A comparative analysis of the outcomes generated by the two QPUs reveals a significant enhancement (reaching nearly a two-fold improvement for PolyFit and a three-fold improvement for VQEC) in the prediction and sampling of the optimal solution (ground state conformations) on the newer Heron R2 architecture, highlighting the impact of quantum hardware advancements for this application. Keywords: quantum computing, protein structure prediction, lattice models 1 Introduction Protein structure prediction (PSP) is a field of interest to many disciplines of life sciences and medicine. Biological activities are often a manifestation of interactions between proteins, functions of which are directly related to the three-dimensional (3D) structures they adopt in specific physiological environments. Understanding fundamental mechanisms of health and disease relies on knowing, or predicting, these structures. Computational models offer an approach to solving this problem. Indeed, the field of template-based modeling (TM), where machine learning algorithms are trained on available experimental structures, has provided some of the most successful PSP methods. The inherent reliance on these experimental structures, which are produced at a much slower rate than the identification of the corresponding primary amino acid sequences (through genomic sequencing), will continue to be a limitation for these methods. Free modeling (FM) methods, in which most of the effort is placed on simulating the physics responsible for yielding these folded structures in nature (such as molecular dynamics simulations and other ab initio PSP methods) continue to be explored as viable alternatives, as they arguably require less of a direct dependence on a database of homologous proteins. Given the computational complexity and scalability limitations of classical FM methods, a growing field has emerged wherein FM approaches are being developed using quantum algorithms (herein referred to as QPSP methods). In most quantum FM methods, conformational spaces are generally discretized onto coarse-grained lattices, and residues are reduced to single spheres centered at each amino acid’s alpha carbon (C \alpha ). Knowledge-based scoring functions, also known as statistical potentials, are used to differentiate conformers. While this inherently implies these are lower resolution models (compared to many explicit all-atom classical FM methods), the obvious benefit is the resulting quantum resource efficiency, which is always a considerable factor especially with pre-fault tolerant quantum hardware. With this in mind, to successfully predict the structure of a protein using a quantum computer, several pieces need to align well. For example, there is a delicate dependence of these models on the construction of the Hamiltonian terms, including the statistical potentials used, and the choice and encoding of a lattice. Optimal solutions are fully determined by the problem Hamiltonian formulations. Aside from this, there are other important components unique to quantum computing, such as the choice of ansatz in variational quantum algorithms (VQAs) for efficient sampling and the classical optimizer chosen. Finding a delicate balance between quantum resources, e.g., qubit count, Hamiltonian locality, etc., and accuracy of the prediction, is a major challenge in the QPSP space. To this end, one of the most resource-efficient QPSP methods is formulated on a tetrahedral lattice [ 1 ] . However, as we will show, this lattice is not very flexible; it is limited in the number of turn angles it can generate, and the number of directions in which the protein chain can grow at each lattice point. Thus, it is not suitable for predicting secondary structures such as \alpha -helices, where the turn angles are notably different from the tetrahedral lattice’s 109.5°. Moreover, there is considerable post-processing needed to generate all-atom biological structures that can be compared to experimental structures to assess model performance. Prior work that utilized QPSP has placed less emphasis on the latter, which is a differentiating factor in our work. Part of our ongoing effort involves optimizing the use of bioinformatics approaches to automate this post-processing, and will be presented in future works. Fig. 1 illustrates how these pieces interconnect to produce a fully realized protein structure prediction. Figure 1: From the selection of an optimal lattice model to final performance analysis, protein structure prediction with quantum algorithms (QPSP) is a complex task that requires a careful analysis of each of these individual pieces. Depicted here is the Trp-cage miniprotein (PDB: 2JOF [ 2 ] ). In this work, we consider a more granular lattice, the face-centered cubic (FCC) lattice, which has not been previously implemented in QPSP. Due to its higher degree of freedom and non-bipartite nature, the encoding of the FCC lattice is more complex than that of the tetrahedral lattice and requires more qubits. We present a quantum encoding of the FCC lattice, focusing on two new methods of constructing the problem Hamiltonian that avoid using slack variables, which are often used to enforce inequality constraints but can lead to an increase in the number of qubits and terms in the problem Hamiltonian. We further analyze the trade-off between the number of qubits and number of terms in the problem Hamiltonian for these methods. Our implementation of the FCC lattice is motivated by our observation that this lattice, compared to earlier coarse-grained lattices, is capable of capturing more realistic biological structures such as secondary structures. Finally, we present our quantum hardware experiment results executed on ibm_cleveland and ibm_kingston , highlighting the improvements on the newer Heron processor. 2 A brief review of coarse-grained lattice models in QPSP The field of QPSP is a relatively young field compared to its classical counterparts. Various coarse-grained lattice models in 2D and 3D for structure prediction such as the cubic and tetrahedral lattices, have been proposed and implemented for quantum computing. The QPSP problem is formulated via discretization of the configuration (or conformational) space, in which the structure prediction is treated as an optimization problem. Finding the optimal structure is then reduced to finding the ground state of a Hamiltonian that encodes the energy of the protein structure. The Hamiltonian is constructed from a set of terms that represent the interactions between the residues in the protein chain, as well as various constraints that ensure the validity of the structure and nearest neighbor pairs. Earlier works explored different lattice models, and various encoding methods to achieve more realistic

Read the complete text