Image licensed by Ingram Publishing

D
Recent Advances
in Quantum
Drug discovery and devel-
opment is a time-consuming and cost-
intensive process. Computer-aided drug
design can speed up the timeline and

Computing for
reduce costs by decreasing the number of
necessary biochemical experiments. The
number of studies using quantum com-
puting to solve problems in drug devel-

Drug Discovery and

opment has been increasing in recent
years. In this review, we briefly intro-
duce the main steps in drug discovery
and development and how computers

Development
help to find potential drug candidates.
Recent studies of quantum computing
in drug development based on the struc-
ture of target proteins are listed chrono-
logically. They include protein structure

Digital Object Identifier 10.1109/MNANO.2023.3249499

Date of current version: 5 May 2023

Pei-Hua Wang, Jen-Hao Chen, Yu-Yuan Yang, Chien Lee, and Yufeng Jane Tseng

Authorized licensed
26 | IEEE use limited to: Birla Institute of Technology & Science. Downloaded on December 02,2023 at 07:34:12 UTC from IEEE Xplore.
nanotechnology magazine | APRIL 2023
Restrictions apply.
1932-4510/12/23©2023IEEE
 prediction, molecular docking, quantum bind to the active site with the highest RESULTS AND DISCUSSION
simulation, and quantitative structure- binding affinity. Quantitative structure- In this section, several quantum com-
activity relationship (QSAR) models. activity relationship (QSAR) prediction puting applications in drug discovery
Current quantum devices are still suscep- models can help to filter out compounds and development will be introduced
tible to noise and error but are well suited with toxicity or other undesired proper- following the timeline described in
for hybrid quantum-classical algorithms. ties. Computer-aided drug development Figure 1. We will first discuss the prob-
The quantum advantage is demonstrated can reduce the number of experiments lem to solve, mention the state-of-art
on hybrid systems and quantum-inspired while maintaining the high probabil- classical methods and finally discuss the
devices such as quantum annealers. We ity of finding safe and effective com- quantum solutions.
hope to see more applications of quan- binations; this decreases the costs and
tum computing in the field of drug dis- time in the drug development [3] and Protein Structure Prediction
covery and development. saves the lives of experimental animals. Protein structure prediction, or protein
Developing new drugs cheaper and faster modeling, is the problem of determining
INTRODUCTION also currently benefits patients with no the structure of a protein given its amino
Drug Development Overview sui treatments. acid sequence. The best traditional mod-
Drug discovery and development typical- eling programs are template-based meth-
ly takes 10-15 years and costs more than Quantum Advantage ods, which use known protein structures
2 billion US dollars for a new drug to Recent developments in quantum com- of similar sequences for initialization
enter the market [1]. The typical process puters enable quantum algorithms to be and optimize the predicted structure
comprises target discovery, hit identifica- implemented and applied to numerous afterward. However, modeling is still
tion, lead optimization, preclinical stag- problems. Computer-aided drug design, tricky when a template is not available.
es, and clinical trials (Figure 1). Target or computational chemistry, is one of Many modern algorithms aiming for
discovery is to find a target protein from the most promising f ields for quan- template-free modeling have been devel-
research that has the potential to cure or tum computing because of its quantum oped. Neural network-based methods for
alleviate certain diseases or symptoms. nature. Determination of the energy protein modeling have achieved remark-
Screening assays are then performed in levels of particle systems is one of the able success in recent years, and they are
a laboratory to see the hit compounds keys to designing pharmaceutical sub- well-suited for novel structure predic-
that can activate or inhibit target proteins stances. Quantum algorithms such as tion [5], [6]. Whether quantum comput-
from thousands of chemical structures. the quantum phase estimation (QPE) or ing can assist in protein modeling has
Lead compounds with more favorable variational quantum eigensolver (VQE) been recently considered. For example,
properties, including absorption, distri- could offer an exponential speedup given Perdomo-Ortiz et al. [7] used quantum
bution, metabolism, excretion, and toxic- robust quantum processing units. Quan- annealing to find low-energy conforma-
ity (ADMET), are identified or modified tum search algorithms and quantum tions of the lattice protein model.
from the hit compounds. The safety and optimization algorithms, such as QAOA, For quantum computers, heuristic
efficacy of lead compounds will first be could also be applied to many optimi- algorithms, such as the quantum approx-
tested with biological molecules and cells zation problems in the pharmaceutical imate optimization algorithm (QAOA)
(in vitro) and animal (in vivo) experi- industry. Furthermore, quantum-assisted [8], [9], provide speedup over classical
ments, followed by clinical trials. machine learning could better represent algorithms. QAOA is a variational algo-
data and build more efficient and accu- rithm for combinatorial optimization
Computer-Aided Drug Design rate models [4]. Quantum computing problems. It is designed to approximately
With the advances in computers, simula- could speed up the drug development minimize a classical cost function depend-
tions and predictions can be performed process further. ing on n binary variables. Fingerhuth
before experiments and trials, reducing
the time and cost spent on compounds
that are unlikely to have the desired
properties. The vast majority of small-
molecule drugs function by binding to
protein targets. [2] Structure prediction
can generate 3D models using a target
protein sequence, even if its structure
is still unknown. The predicted protein
structure can be used in virtual screen-
ing to discover compounds with a high
potential to activate or inhibit target pro-
teins. For example, in prediction, dock-
ing studies can find compounds that FigurE 1 Overview of computer-aided drug development.

Authorized

licensed use limited to: Birla Institute of Technology & Science. Downloaded on December 02,2023 at 07:34:12 UTC from IEEE Xplore. Restrictions apply.
APRIL 2023 | IEEE nanotechnology magazine | 27
 et al. [10] used a novel encoding method a known ligand-protein complex [14]. VQE [21] is a variational method that
for one-hot encoded turns of a 3D lat- The docking problem was mapped to varies the state vector ansatz to obtain the
tice protein folding problem. The novel find the maximum weighted clique from minimum eigenvalue of a given Hamilto-
encoding method produces Hamiltoni- the binding interaction graph, including nian. Its hybrid nature, iterative execu-
ans that can be minimized with QAOA vertices of pharmacophores and edges of tion of relatively short quantum circuits,
circuits. The models can explore many interactions between the pharmacoph- and classical optimization make it suit-
possible protein conformations and ores. able to run on current noisy intermedi-
infer the structural properties of more Molecular unfolding, a procedure in ate-scale quantum (NISQ) devices. After
complex atomistic protein structures. molecular docking to relax the ligand mapping the system Hamiltonian to a
Boulebnane et al. [11] investigated the structure and remove initial bias, was qubit representation with second quan-
feasibility of sampling low-energy lattice- performed on D-wave annealing hard- tization and fermion-qubit mappings,
based peptides using QAOA. They ware in Mato's study [15]. To identify such as the Jordan-Wigner transforma-
showed promising results in sampling a the best configuration of a molecule, tion, VQE utilizes traditional optimi-
self-avoiding walk while they struggled the sum of the distances between atoms zation algorithms to update parameters
to find low-energy instances for a small comprising a molecule is expressed as to prepare the ansatz by viewing the
peptide. They suggested that deep quan- quadratic unconstrained binary optimi- execution result of the parametrized cir-
tum circuits are required to achieve accu- zation (QUBO) problem and is maxi- cuit as a cost function to be minimized.
rate results. mized; the angles of rotatable bonds Although VQE has been successfully
Grover's search algorithm offers are taken as parameters. The outcomes applied to the energy determination of
a quadruple speedup and widespread and performance of molecular unfold- systems, such as water molecule [22]
applicability in search-based method- ing using quantum annealing are com- and hydrogen chain systems [23], it still
ologies. Wong et al. [12] developed a parable to the state-of-the-art classical requires further advancement to be gen-
quantum algorithm with a Grover search algorithms in this study. erally applied. For example, VQE also
subroutine in a hydrophobic-hydrophilic suffers from a large circuit size and many
model on a 2D square lattice to solve the Quantum Simulation ansatz parameters as the size of the sys-
problem for any sequence of N amino Determining the energy of a multiparti- tem of interest increases. In addition,
acids with a quadruple speedup over its cle system, such as a protein-ligand com- there is no fast convergence guarantee
classical counterpart. plex, requires solving the Schrödinger for optimizing the parameters.
equation, which becomes intractable
Molecular Docking when the number of particles grows. Quantitative Structure-Activity
Molecular docking is a computational Simulation algorithms have been devel- Relationship (QSAR)
tool often used in the next step to find oped to approximate the solution. The PREDICTION Model
promising hit compounds when protein molecular dynamics (MD) [16] evolves A quantitative structure-activity rela-
structure becomes available. In virtu- a system using classical equations with tionship (QSAR) is an in silico method
al screening, compound libraries that small time intervals, whereas Monte- based on known databases of structures
include known drug-like molecules are Carlo [17] methods randomly sample and activity to predict a novel structure's
evaluated using molecular docking to system configurations under some ener- physiological activity. Scientists must per-
identify those that can bind to a target gy criteria. Moreover, density functional form many time-consuming and expen-
protein's active or ligand-binding site. theory (DFT) [18] methods are preferred sive biochemical experiments to acquire
For example, PanGPCR [13] is a system at the subatomic scale because they con- physiological activity. Moreover, some
that dock the compound of interest to 46 sider electronic interactions in a system, experiments require model organisms,
known docking sites for human G pro- where the Hamiltonian of the system such as mice or rabbits, for testing. Seek-
tein-coupled receptors (GPCRs) to pre- is approximated under certain assump- ing alternatives to animal testing is an
dict potential GPCR target, expression tions. For example, Tu et al. [19] utilized important issue. A well-designed model
location, and side effects. The best dock- the Becke, three-parameter, LeeYang- can provide reasonable estimations of
ing poses of a molecule can be predicted Parr (B3LYP) density functional to cal- chemical activities.
by maximizing docking scores derived culate the thermodynamic properties of Zhang et al. [24] summarized several
from favorable noncovalent interactions, alternariol-related chemical structures commonly used machine learning meth-
such as hydrogen bonds, electrostatic and simulate the transition of tauto- ods, including support vector machine
interactions, or hydrophobic interactions mers. Simulation algorithms, however, (SVM), decision tree (DT)[25], random
between a ligand and the target protein. can be inaccurate and fail dramatically forest (RF), k nearest neighbor (kNN),
Docking studies that utilize quan- in some situations. Acquiring the exact and artificial neural network (ANN) (com-
tum devices have recently emerged. Ban- energy under full configuration interac- monly classified as deep learning approach-
chi et al. used Gaussian boson samplers, tion (FCI), which considers all the sys- es) methods, that are used in QSAR
nonuniversal photonic quantum devices, tem's electrons, still requires exponential models. Deep learning methods, such
to correctly predict the docking pose of resources [20] in classical computing. as convolutional neural networks (CNN

Authorized licensed
28 | IEEE use limited to: Birla Institute of Technology & Science. Downloaded on December 02,2023 at 07:34:12 UTC from IEEE Xplore. Restrictions apply.
nanotechnology magazine | APRIL 2023
 s) [26] and recurrent neural networks provided a procedure for structure- Universal Quantum Computer
(RNNs), are often used as QSAR models. based data, the optimal kernel design of and Quantum-Inspired Methods
In Yu's study [27], ensemble models com- QSVM is an open question for further The number of qubits and the depth of
bining machine learning and deep learn- investigations [36], [37], [38], [39]. Last, the circuits on current NISQ computers
ing can achieve high accuracy in predicting a quantum algorithm for QSAR tasks are limited, thus limiting the scale of the
CNS accessibility and provide interpretable has yet to be well designed. problems. The examples are simplified
generalized rules for simple classification. mainly to run on the actual device for
In the general framework, the chemi- Others proof-of-concept experiments. On the
cal features are extracted from the molec- Several studies are worth mentioning other hand, the quantum advantage of
ular structures by third-party platforms, in addition to the standard workf low hybrid quantum-classical algorithms was
such as RDKit [28]. Users can choose of structure-based drug development, demonstrated in drug development since
whether to use dimensional reduction which utilizes the structures of both the parametrized quantum circuits with
techniques before performing machine target protein and the ligands as intro- short depths are sufficient for calculating
learning or deep learning, depending on duced above. Patent search is a critical parameters in the next step. We can also
the requirements. On the other hand, a step to avoid infringement before apply- see that annealers are better at solving
molecular structure can be transformed ing for pharmaceutical composition pat- several problems that can be reduced to
into a text string using simplified molec- ents. Ligand-based drug design explores combinatorial optimization problems.
ular-input line-entry system (SMILES) a chemical space to find new structurally
notation. Though one canonical chemi- similar compounds with the potential to CONCLUSION
cal structure is often used to represent become drugs according to the ligands Quantum computation is starting to
one chemical structure, complete enu- with known activity. attract attention in the field of drug devel-
meration of SMILES notations improves Wang et al. [40] proposed a Grover opment. We overviewed the typical drug
the performance of the CNN predic- search-based method that performs an development process and how comput-
tion model for aqueous solubility [29]. exact comparison between two non-enu- ers help accelerate the timeline. Then we
A graph that denotes a molecular struc- merated Markush structures with a con- introduced the problems in drug devel-
ture can be regarded as input for the straint satisfaction oracle. The quantum opment and solutions using quantum
graph convolution neural network method they proposed compares two computing, including protein structure
(GCN) QSAR model [30]. patents that are hard to simulate classi- prediction, molecular docking, quantum
A quantum SVM (QSVM) model is cally and thus creates a quantum advan- simulation, QSAR prediction model,
a hybrid approach that involves quan- tage in patent analysis. patent search, and ligand-based drug
tum and classical methods [31]. It Hernandez et al. [41] introduced a design. Combining the knowledge in
replaces traditional kernels with quan- quantum-inspired graph-based molecu- bioinformatics, cheminformatics, medici-
tum kernels, such as linear or radial basis lar similarity (GMS) method for ligand- nal chemistry, and quantum computing,
function kernels. QSVM may provide based virtual screening. The virtual pharmaceutical problems that are hard for
prospective quantum advantages in the screening results of GMS, including 3D classical computers are becoming solvable
drug discovery [32]. Batra et al. applied atomic coordinates as features, outper- if using quantum computers with more
QSVM in QSAR tasks with several formed conventional fingerprints by qubits and less noise. There will be more
bioactivity datasets and found a slight finding more active compounds in the cross-disciplinary studies to solve drug
running time advantage on a quantum top few percent of the output list. development problems using quantum
computer simulator compared to a classi- In Li's study [42], a scalable quantum devices or quantum-inspired methods.
cal computer [33]. To implement QSVM generative autoencoder is presented and
on a real QSAR problem, the classical used to generate novel molecules taking ACKNOWLEDGMENTs
data should be efficiently encoded into available drugs as input. An autoencoder Resources at the Laboratory of Compu-
current quantum devices. Mensa et al. comprises an encoder and a decoder. The tational Molecular Design and Metab-
proposed a general framework for the encoder is trained to encode the input olomics, the Department of Computer
ligand-based virtual screening (LB-VS) parameters to the latent space. The decod- Science and Information Engineering at
task [32], which can be generalized to er is trained to reconstruct the informa- National Taiwan University, and Phys-
solve problems in cheminformatics. tion from the latent space to the original ics Division, National Center for Theo-
These high-dimensional data were input. If noise is introduced into the latent retical Sciences were used in performing
not available for near-term quantum space, the autoencoder will generate novel this study. This work was supported in
devices; thus, classical skills of dimen- outputs similar to the inputs. The quan- part by the National Science and Tech-
sional reduction were needed, i.e., prin- tum autoencoder utilizes the probabilistic nology Council, under MOST Grants
cipal component analysis (PCA) [34] nature of quantum circuits as noise and 111-232 0 -B - 0 02 - 0 43 -M Y 2 , 111-
or selecting some best features with generates molecules with better drug 2926-I-002-501, and 111-2119-M-
the analysis of variance (ANOVA) [35]. properties than its classical counterpart 033-001, in part by Taiwan Food and
Furthermore, although the framework with high-dimensional latent space. Drug Administration under Grants

Authorized

licensed use limited to: Birla Institute of Technology & Science. Downloaded on December 02,2023 at 07:34:12 UTC from IEEE Xplore. Restrictions apply.
APRIL 2023 | IEEE nanotechnology magazine | 29
 MOHW111-FDA-D-114-000611 and [2] Y. Shan, V. P. Mysore, A. E. Leffler, E. T. Kim, S. computer,” NPJ Quantum Inf., vol. 6, no. 1,
Sagawa, and D. E. Shaw, “How does a small mol- pp. 1–6, 2020.
MOH W112-FDA-D-114-000611, in ecule bind at a cryptic binding site?,” PLoS Com- [23] G. A. Quantum et al., “Hartree-Fock on a super-
part by the National Taiwan University put. Biol., vol. 18, no. 3, 2022, Art. no. e1009817. conducting qubit quantum computer,” Science,
[3] S. J. Y. Macalino, V. Gosu, S. Hong, and S. vol. 369, no. 6507, pp. 1084–1089, 2020.
Higher Education Sprout Project under Choi, “Role of computer-aided drug design in [24] L. Zhang, J. Tan, D. Han, and H. Zhu, “From
Grant NTU-111L8809 and NTU-CC- modern drug discovery,” Arch. Pharmacal. Res., machine learning to deep learning: Progress
vol. 38, pp. 1686–1701, 2015. in machine intelligence for rational drug dis-
111L890203, and in part by Toxic and [4] C. Outeiral, M. Strahm, J. Shi, G. M. Morris, S. covery,” Drug Discov. Today, vol. 22, no. 11,
Chemical Substances Bureau, Environ- C. Benjamin, and C. M. Deane, “The prospects pp. 1680–1685, 2017.
of quantum computing in computational molec- [25] P.-H. Wang, Y.-S. Tu, and Y. J. Tseng, “PgpRules:
mental Protection Administration, Execu- ular biology,” Wiley Interdiscipl. Rev.: Comput. A decision tree based prediction server for P-
tive Yuan, under Grant 110A022. Mol. Sci., vol. 11, no. 1, 2021, Art. no. e1481. glycoprotein substrates and inhibitors,” Bioin-
[5] J. Jumper et al., “Highly accurate protein formatics, vol. 35, no. 20, pp. 4193–4195, 2019.
structure prediction with AlphaFold,” Nature, [26] J.-H. Chen and Y. J. Tseng, “A general opti-
ABOUT THE AUTHORS vol. 596, no. 7873, pp. 583–589, 2021. mization protocol for molecular property pre-
[6] M. Baek et al., “Accurate prediction of protein diction using a deep learning network,” Brief.
Pei-Hua Wang (phwang1231@ structures and interactions using a three-track Bioinf., vol. 23, no. 1, 2022, Art. no. bbab367.
cmdm.csie.ntu.edu.tw) is with the Quan- neural network,” Science, vol. 373, no. 6557, [27] T.-H. Yu, B.-H. Su, L. C. Battalora, S. Liu, and
pp. 871–876, 2021. Y. J. Tseng, “Ensemble modeling with machine
tum Information Center, Chung Yuan [7] A. Perdomo-Ortiz, N. Dickson, M. Drew- learning and deep learning to provide inter-
Christian University, Taoyuan City, Brook, G. Rose, and A. Aspuru-Guzik, “Find- pretable generalized rules for classifying CNS
ing low-energy conformations of lattice protein drugs with high prediction power,” Brief. Bio-
320314, Taiwan and also wtih Graduate models by quantum annealing,” Sci. Rep., vol. 2, inf., vol. 23, no. 1, 2022, Art. no. bbab377.
Institute of Biomedical Electronics and no. 1, pp. 1–7, 2012. [28] “G. Landrum, RDKit: Open-source chemin-
[8] E. Farhi, J. Goldstone, and S. Gutmann, “A formatics 2010,” [Online]. Available: https://
Bioinformatics, College of Electrical Engi- quantum approximate optimization algorithm,” www.rdkit.org.
neering and Computer Science, National 2014, arXiv:1411.4028. [29] J.-H. Chen and Y. J. Tseng, “Different molecu-
[9] S. Hadfield, Z. Wang, B. O'gorman, E. G. Rief- lar enumeration influences in deep learning: An
Taiwan University, Taipei, 10617, Taiwan. fel, D. Venturelli, and R. Biswas, “From the example using aqueous solubility,” Brief. Bioinf.,
Jen-Hao Chen ([email protected]. quantum approximate optimization algorithm vol. 22, no. 3, 2021, Art. no. bbaa092.
to a quantum alternating operator ansatz,” [30] A. Renn, B. H. Su, H. Liu, J. Sun, and Y. J.
ntu.edu.tw) is with the Chunghwa Tele- Algorithms, vol. 12, no. 2, 2019, Art. no. 34. Tseng, “Advances in the prediction of mouse liver
com Co., Ltd., Taipei, Taiwan, and also [10] M. Fingerhuth and T. Babej, “A quantum microsomal studies: From machine learning to
alternating operator ansatz with hard and soft deep learning,” Wiley Interdiscipl. Rev.: Comput.
with Department of Computer Science constraints for lattice protein folding,” 2018, Mol. Sci., vol. 11, no. 1, 2021, Art. no. e1479.
and Information Engineering, National arXiv:1810.13411. [31] D. P. García, J. Cruz-Benito, and F. J. García-
[11] S. Boulebnane, X. Lucas, A. Meyder, S. Peñalvo, “Systematic Literature Review: Quan-
Taiwan University, Taipei, 10617, Taiwan. Adaszewski, and A. Montanaro, “Peptide tum Machine Learning and its applications,”
Yu-Yuan Yang ([email protected]. conformational sampling using the Quantum 2022, arXiv:2201.04093.
Approximate Optimization Algorithm,” 2022. [32] S. Mensa, E. Sahin, F. Tacchino, P. K. Barkoutsos,
edu.tw) is with the Graduate Institute of arXiv:2204.01821. and I. Tavernelli, “Quantum Machine learning
Biomedical Electronics and Bioinformat- [12] R. Wong and W.-L. Chang, “Fast quantum framework for virtual screening in drug discovery:
algorithm for protein structure prediction in A prospective Quantum advantage,” Mach. Learn.:
ics, College of Electrical Engineering hydrophobic-hydrophilic model,” J. Parallel Dis- Sci. Technol., vol. 4, no. 1, 2023, Art. no. 015023.
and Computer Science, National Taiwan trib. Comput., vol. 164, pp. 178–190, 2022. [33] K. Batra et al., “Quantum machine learn-
[13] L.-C. Liu, M.-Y. Ho, B.-H. Su, S.-Y. Wang, ing algorithms for drug discovery applica-
University, Taipei, 10617, Taiwan. M.-T. Hsu, and Y. J. Tseng, “PanGPCR: Pre- tions,” J. Chem. Inf. Model., vol. 61, no. 6,
Chien Lee (chienlee1003@cmdm. dictions for multiple targets, repurposing and pp. 2641–2647, 2021.
side effects,” Bioinformatics, vol. 37, no. 8, [34] S. Wold, K. Esbensen, and P. Geladi, “Principal
csie.ntu.edu.tw) is with the Department pp. 1184–1186, 2021. component analysis,” Chemometrics Intell. Lab.
of Computer Science and Information [14] L. Banchi, M. Fingerhuth, T. Babej, C. Ing, and Syst., vol. 2, no. 1-3, pp. 37–52, 1987.
Engineering, National Taiwan Univer- J. M. Arrazola, “Molecular docking with Gauss- [35] L. St and S. Wold, “A nalysis of variance
ian boson sampling,” Sci. Adv., vol. 6, no. 23, (ANOVA),” Chemometrics Intell. Lab. Syst.,
sity, Taipei, 10617, Taiwan. 2020, Art. no. eaax1950. vol. 6, no. 4, pp. 259–272, 1989.
Yufeng Jane Tseng (correspond- [15] K. Mato, R. Mengoni, D. Ottaviani, and G. Pal- [36] V. Rastunkov et al., “Boosting method for
ermo, “Quantum molecular unfolding,” Quan- automated feature space discovery in supervised
ing author: ([email protected])) is tum Sci. Technol., vol. 7, 2022, Art. no. 035020. quantum machine learning models,” 2022,
with the Graduate Institute of Biomedical [16] M. Sprik and G. Ciccotti, “Free energy from arXiv:2205.12199.
constrained molecular dynamics,” J. Chem. [37] S. Altares-López, A. Ribeiro, and J. J. García-
Electronics and Bioinformatics, College of Phys., vol. 109, no. 18, pp. 7737–7744, 1998. Ripoll, “Automatic design of quantum feature
Electrical Engineering and Computer Sci- [17] G. M. Torrie and J. P. Valleau, “Nonphysical maps,” Quantum Sci. Technol., vol. 6, no. 4,
sampling distributions in Monte Carlo free- 2021, Art. no. 045015.
ence, National Taiwan University, Taipei, energy estimation: Umbrella sampling,” J. Com- [38] E. Torabian and R. V. Krems, “Optimal quan-
10617, Taiwan, and also with Depart- put. Phys., vol. 23, no. 2, pp. 187–199, 1977. tum kernels for small data classification,” 2022,
[18] W. Kohn and L. J. Sham, “Self-consistent arXiv:2203.13848.
ment of Computer Science and Infor- equations including exchange and correlation [39] E. Peters et al., “Machine learning of high
mation Engineering, National Taiwan effects,” Phys. Rev., vol. 140, no. 4A, 1965, dimensional data on a noisy quantum processor,”
Art. no. A1133. NPJ Quantum Inf., vol. 7, no. 1, pp. 1–5, 2021.
University, Taipei, 10617, Taiwan, and [19] Y.-S. Tu, Y. J. Tseng, and M. Appell, “Quan- [40] P.-H. Wang, J.-H. Chen, and Y. J. Tseng, “Intel-
also with Physics Division, National Cen- tum chemical investigation of the detection ligent pharmaceutical patent search on a near-
properties of alternariol and alternariol mono- term gate-based quantum computer,” Sci. Rep.,
ter for Theoretical Sciences, Taipei, 10617, methyl ether,” Struct. Chem., vol. 30, no. 5, vol. 12, no. 1, pp. 1–8, 2022.
Taiwan, and also with Center for Quan- pp. 1749–1759, 2019. [41] M. Hernandez, G. Liang Gan, K. Linvill, C.
[20] A. Szabo and N. S. Ostlund, Modern Quantum Dukatz, J. Feng, and G. Bhisetti, “A quantum-
tum Science and Engineering, National Chemistry: Introduction to Advanced Electronic inspired method for three-dimensional ligand-
Taiwan University, Taipei, 10617, Taiwan. Structure Theory. Chelmsford, CA, USA: Cou- based virtual screening,” J. Chem. Inf. Model.,
rier, 2012. vol. 59, no. 10, pp. 4475–4485, 2019.
[21] A. Peruzzo et al., “A variational eigenvalue solv- [42] J. Li and S. Ghosh, “Scalable variational quan-
REFERENCES er on a photonic quantum processor,” Nature tum circuits for autoencoder-based drug discov-
[1] N. Berdigaliyev and M. Aljofan, “An overview of Commun., vol. 5, no. 1, pp. 1–7, 2014. ery,” in Proc. IEEE Des., Automat. Test Europe
drug discovery and development,” Future Med. [22] Y. Nam et al., “Ground-state energy estimation Conf. Exhib., 2022, pp. 340–345.
Chem., vol. 12, no. 10, pp. 939–947, 2020. of the water molecule on a trapped-ion quantum 

Authorized licensed
30 | IEEE use limited to: Birla Institute of Technology & Science. Downloaded on December 02,2023 at 07:34:12 UTC from IEEE Xplore. Restrictions apply.
nanotechnology magazine | APRIL 2023