Leveraging Machine Learning to Overcome Limitations in Quantum Algorithms
Laia Coronas Sala 1, Parfait Atchade-Adelemou 2,3 · 2024
Read Leveraging Machine Learning to Overcome Limitations in Quantum Algorithms by Laia Coronas Sala 1, Parfait Atchade-Adelemou 2,3 (2024) 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-2412-11405.txt.
Introduction
Laia Coronas Sala and Parfait Atchade-Adelemou blend machine learning with quantum computing for drug research in a 2024 paper. Quantum methods promise strong tools for describing molecules, above all their ground state energy, a key value for judging drug candidates. Today's quantum machines, though, suffer noise, small size, and complex setups.
Their hybrid frame pairs learning models with three quantum methods, from a variational solver to phase estimation. They train on three data sets built from public molecule records, mixing chemical traits and charge maps. The learning side helps correct and steady the quantum results. The work points toward practical molecule screening on near-term quantum hardware, where pure quantum methods still fall short.
Catalog record
| CREATOR | Laia Coronas Sala 1, Parfait Atchade-Adelemou 2,3 |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-11405
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Leveraging Machine Learning to Overcome Limitations in Quantum Algorithms Laia Coronas Sala 1, Parfait Atchade-Adelemou 2,3 Abstract Abstract Quantum Computing (QC) offers outstanding potential for molecular characterization and drug discovery, particularly in solving complex properties like the Ground State Energy (GSE) of biomolecules. However, QC is constrained by computational noise, scalability, and system complexity. This work presents a hybrid framework that combines Machine Learning (ML) techniques with quantum algorithms, namely the Variational Quantum Eigensolver (VQE), Hartree-Fock (HF), and Quantum Phase Estimation (QPE)—to improve GSE predictions when computing large molecules. We curated three datasets—chemical descriptors, Coulomb matrices, and a hybrid combination—using molecular features from PubChem. These datasets were used to train XGBoost (XGB), Random Forest (RF), and LightGBM (LGBM) models. Results show that XGB achieved the lowest Relative Error (RE) of 4.41 ± 11.18% on chemical descriptors, outperforming RF (5.56 ± 11.66%) and LGBM (5.32 ± 12.87%). The HF method delivered exceptional precision for small molecules (0.44 ± 0.66% RE), while a near-linear correlation between GSE and molecular electron count provided predictive shortcuts. This study demonstrates that integrating QC and ML enhances scalability for molecular energy predictions. The proposed approach lays the foundation for overcoming QC’s limitations and scaling molecular simulations to larger systems. keywords Quantum Mechanics, Quantum Chemistry, Molecular Biology, Amino Acids, Schrödinger Equation, Hamiltonian Operator, Ground State Energy, Coulomb Matrices, Chemical Properties, Machine Learning, SHAP Values 1 University of Barcelona, Biomedical Engineering, Barcelona, Spain. Email: [email protected] 2 Lighthouse Disruptive Innovation Group S.L, Barcelona, Spain. Email: [email protected] 3 MIT Media Lab, City Science, Boston, MA, USA. Email: [email protected] 1 INTRODUCTION Accurate protein and drug molecule characterization has driven the development of diverse computational and experimental techniques [ 1 ] . Since John Dalton’s 19 th -century atomic theory, it is known that molecules comprise atoms made of electrons, protons, and neutrons [ 2 ] . This raises a critical question: Can deeper insights into subatomic behavior enhance molecular characterization? While modern tools like AlphaFold, Rosetta, and I-TASSER have revolutionized the prediction of three-dimensional structures [ 3 , 4 , 5 ] , they rely on approximations that can limit accuracy, increasing costs and timelines in drug development [ 6 , 7 , 8 , 9 ] . Quantum mechanics, by contrast, explores particle behavior at subatomic scales, capturing interactions and energy states with unparalleled precision. Unlike homology-dependent classical methods, quantum chemistry is grounded in physical principles of energy and entropy, making it ideal for high-precision conformational predictions [ 10 , 11 ] . Key quantum properties such as superposition enable quantum computers to process multiple states simultaneously, vastly outperforming classical systems in complex computations [ 12 ] . Quantum mechanics models biological systems via wave functions, solutions to the Schrödinger equation ( 1 ), which predict electron behavior and molecular properties. i\hbar\frac{\partial}{\partial t}|\psi\rangle=\hat{H}|\psi\rangle, (1) The Hamiltonian operator \hat{H} , Eq:( 2 ), represents system energy, simplified by the Born-Oppenheimer approximation [ 13 ] . \hat{H}=-\sum_{i=1}^{N}\frac{1}{2}\nabla_{i}^{2}-\sum_{i=1}^{N}\sum_{A=1}^{M}\frac{Z_{A}}{r_{iA}}+\sum_{j>i}\frac{1}{r_{ij}}, (2) Where \hat{H} represents the Hamiltonian operator, N is the total number of electrons in the system, \nabla_{i}^{2} is the Laplacian operator acting on the i -th electron, M is the total number of nuclei in the system, Z_{A} is the atomic number of the A -th nucleus representing the charge of the nucleus, r_{iA} is the distance between the i -th electron and the A -th nucleus, and r_{ij} is the distance between the i -th and j -th electrons. Solving the time-independent Schrödinger equation ( 3 ) yields the eigenvalues E , representing allowed energy states: \hat{H}|\psi\rangle=E|\psi\rangle, (3) The lowest eigenvalue corresponds to the Ground State Energy (GSE), the system’s most stable configuration [ 14 ] . This state underpins molecular stability, reactivity, and interaction dynamics, informing reaction pathways and drug binding processes [ 15 ] . However, solving such equations for relatively small systems is already very challenging since it involves solving high search space equations. Therefore, developing efficient and scalable procedures to compute the GSE for large molecules remains an open challenge. 1.1 Motivation The motivation for this work stems from the challenges of finding exact solutions for quantum equations in biological systems with over 30 electrons due to the computational complexity of quantum simulations [ 16 ] . While awaiting advancements in hardware and microelectronics, this project explores strategies to mitigate qubit limitations by integrating quantum methods, datasets, and ML for analyzing complex biomolecules. To test this approach, we focused on computing the ground-state energies of the 20 essential amino acids found in the human body due to their profound biological relevance [ 17 ] . As the building blocks of proteins, these amino acids play a central role in numerous biochemical processes, making them ideal candidates for exploring molecular interactions and dynamics while highlighting the practical implications of this research. Furthermore, this strategy aims to provide the necessary tools for scaling up to larger molecular systems, such as small proteins. With this ultimate objective, we set the following pipeline: • Develop a pipeline to assess the GSEs, examining different Hamiltonian encoders (Jordan-Wigner and Bravyi-Kitaev [ 18 , 19 , 20 ] ) and quantum strategies (Quantum Phase Estimation [QPE] [ 21 ] , Variational Quantum Eigensolver [VQE] [ 22 , 23 , 24 , 25 ] , and the Hartree-Fock Approximation [HF] [ 26 , 27 , 28 ] ). • Create robust training datasets incorporating various molecular descriptors to train ML models and identify feature combinations that yield optimal accuracies. • Build ML models trained on these datasets to predict GSEs for larger molecules that cannot be directly simulated using QC, focusing on the 20 essential amino acids. 2 RELATED WORK The first step before any simulation can begin is encoding the Hamiltonian. For that, the Jordan-Wigner transformation has shown efficacy in calculating the chemical properties of small molecules [ 29 , 18 ] . However, its long sequences of ZZ operations can significantly increase resource demands on quantum hardware [ 30 ] . The Bravyi-Kitaev transformation offers an alternative that optimizes qubit usage and enhances computational efficiency while maintaining accuracy [ 31 , 32 , 33 ] . Once encoded, quantum strategies like QPE and VQE have been effective for extracting GSEs [ 34 , 35 , 36 ] . These quantum methods have been applied to small, computationally feasible molecules, however, for larger molecules, Machine learning (ML) has emerged as an alternative to extrapolate quantum properties to more complex systems that cannot be easily simulated. For example, Dominic et al. and Lauv et al. demonstrated promising results using ML models for generating drug-like molecules [ 37 , 38 ] , supported by similar findings from other studies [ 39 , 40 , 41 ] . Recognizing the importance of robust training datasets, efforts have been made to construct comprehensive datasets. For instance, QM7-X, containing 4.2 million small organic molecules, has proven valuable for predicting ground state properties [ 42 ] . Predecessors like QM8 and QM9 have also contributed significantly [ 43 , 44 , 45 ] , alongside QMugs, which focuses on drug-like molecules [ 46 ] . However, the need for larger, more diverse datasets persists, with some advocating for the inclusion of reoriented molecules to increase training samples and enhance accuracy [ 47