Pith. sign in

REVIEW 3 major objections 4 minor 21 references

Quantum-Enhanced Classification of Brain Tumors Using DNA Microarray Gene Expression Profiles

T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read A 15-qubit variational quantum classifier labels four brain tumor types and healthy tissue from gene expression with about 85% accuracy.

desk verdict The full-feature result is uninstantiable: 15 qubits cannot amplitude-encode 54,676 features, and the classical comparison does not support 'quantum-enhanced'. read the letter →

arxiv 2505.02033 v1 pith:LAGVK7C3 submitted 2025-05-04 cs.LG q-bio.GNq-bio.MN

classification cs.LGq-bio.GNq-bio.MN
keywords quantummachinelearningvariationalclassifierbraintumorclassificationDNAmicroarraygeneexpressionamplitudeencodinghardwareefficientansatzNISQ
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper aims to show that a variational quantum classifier, named Deep VQC, can sort DNA microarray samples into five classes—four brain tumor types and healthy tissue—with useful accuracy. This matters because gene-expression data are extremely high-dimensional, with 54,676 features per sample, and classical methods face growing computational costs in this regime. The authors report validation accuracy of 0.79 when all 54,676 features are fed directly into the model, 0.86 after principal component analysis reduces the input to 65 dimensions, and an average of 0.85 under three-fold cross-validation. They also argue the quantum model is competitive with classical machine-learning baselines on the same dataset.

What carries the argument

The workhorse is the Deep VQC, built from amplitude encoding, two hardware-efficient ansatze, and a softmax readout. Amplitude encoding expresses a normalized classical vector $x$ of length $2^n$ as the amplitudes of an $n$-qubit state, so a 15-qubit circuit can in principle carry at most $2^{15}=32{,}768$ amplitude components. The first HEA applies Hadamard gates, single-qubit RX and RY rotations, and CNOT and Toffoli entangling gates; the second uses Hadamard, RY and RZ rotations with the same entanglers. Measurement in the Pauli-Z basis on five qubits yields class probabilities through the softmax in Eq. (3), and the cross-entropy cost in Eq. (4) is minimized by gradient descent. This combination is what lets the model learn correlations among the gene-expression features while keeping the circuit shallow enough for NISQ hardware.

What would settle it

Write out the 15-qubit state vector: amplitude encoding accepts at most $2^{15}=32{,}768$ amplitudes. Trying to load a 54,676-dimensional vector therefore fails unless the input is truncated, padded, or otherwise reduced; checking how the reported full-feature run handled this settles whether the result is reproducible. Re-running the classical baselines on the same three-fold splits would settle whether the accuracy comparison is fair.

Watch

Extended reading notes

Core claim

The paper's central claim is that its Deep VQC model, a variational quantum classifier with 15 qubits and 25 layers, can separate five classes—ependymoma, glioblastoma, medulloblastoma, pilocytic astrocytoma, and healthy samples—from DNA microarray gene-expression data. The authors report validation accuracy 0.79 when all 54,676 features are transferred into the quantum model, 0.86 after principal component analysis reduces the input to 65 dimensions, and 0.85 average accuracy under three-fold cross-validation. They compare these numbers with classical machine-learning baselines and find the quantum model matches decision trees and naive Bayes, beats a multilayer perceptron, and trails support vector machines and random forests. On the paper's own account, the result is evidence that quantum AI can give competitive or better classification on high-dimensional biological data in the NISQ era.

Load-bearing premise

The whole result rests on the assumption that all 54,676 gene expression values can be loaded directly into a 15-qubit amplitude-encoding circuit, which the paper's own encoding formula does not allow at that size.

Editorial extensions

If this is right

  • If the reported accuracy holds, a variational classifier with only 15 qubits is enough for a five-class medical diagnosis task on gene-expression data, so quantum classifiers are within reach of current hardware.
  • The gain from PCA (validation accuracy 0.86 over 0.79 on full features) suggests aggressive but variance-preserving dimensionality reduction can help a quantum classifier on high-dimensional biological data.
  • Matching decision trees and naive Bayes while beating a multilayer perceptron implies quantum variational classifiers are not automatically worse than classical methods on this kind of task.
  • The 25-layer two-ansatz structure indicates that circuit depth, not just qubit count, can be used as a resource in variational classifiers for microarray data.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The encoding formula in the paper accepts at most $2^{15}=32{,}768$ amplitudes, so the reported run with 54,676 features implies some unstated truncation, padding, or feature-reduction step; identifying that step would make the full-feature result reproducible.
  • The classical comparison is meaningful only if the same cross-validation splits and preprocessing were used; the paper does not document them, so the ranking against decision trees and naive Bayes should be read with that caveat.
  • The same two-ansatz Deep VQC design could be tried on other high-dimensional biological data, such as RNA-seq or methylation arrays, to test whether the pattern generalizes beyond DNA microarrays.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes a hybrid quantum-classical classifier called Deep VQC, which uses amplitude encoding, two hardware-efficient ansatze, and 15 qubits to classify five classes (four brain tumor types plus healthy samples) from CuMiDa microarray data with 54,676 gene features. The authors report a validation accuracy of 0.79 for the full-feature run, 0.86 when PCA is applied (reducing to 65 dimensions), and an average 3-fold cross-validation accuracy of 0.85. They compare these results with published CuMiDa classical ML baselines and conclude that the quantum model achieves competitive or superior performance.

Significance. If the claims were sound, the paper would provide a useful data point on applying variational quantum classifiers to high-dimensional genomic data, with a clear description of a circuit architecture and use of a public benchmark. The paper honestly reports learning curves and confusion matrices, and it cites the public CuMiDa database, which supports reproducibility of the dataset. However, the central claim is undermined by a direct encoding capacity mismatch: the full-feature experiment as described cannot be implemented on the stated 15-qubit circuit. In addition, the classical comparison uses published scores from a different protocol rather than same-protocol runs, and no error bars or repeated-run statistics are provided. These issues make the headline accuracy claims unsupported and the 'quantum-enhanced' framing an overreach.

major comments (3)
  1. [Section II.B.1, Eq. (1); Section II.A.2; Section III.A] The full-feature experiment is not executable as described. Eq. (1) defines amplitude encoding for an input vector of length 2^n, so with n=15 the state space has only 2^15 = 32,768 amplitude slots. Section II.A.2 states that normalized data were 'directly fed' into the Deep VQC model 'while preserving the original 54,676 gene features,' and Section III.A reiterates that all 54,676 features were transferred to the 15-qubit model. No truncation, feature selection, padding, or alternative encoding is described for the full-feature run. Therefore the reported 0.79 validation accuracy for this configuration cannot be reproduced on the stated hardware, and the abstract's '54,676 gene features' claim is unsupported.
  2. [Section III.C and Table I] The comparison with classical ML models is not made under a common protocol. Table I reports CuMiDa database accuracy scores for classical algorithms, while the Deep VQC result comes from the authors' own 3-fold cross-validation run with a different (and unspecified) split, preprocessing, and hyperparameter selection. Without identical train/test partitions, feature scaling, and evaluation procedure, the accuracy values are not directly comparable. Moreover, even taking Table I at face value, the quantum model's 0.85 is lower than SVM (0.95), RF (0.91), and KNN (0.87), which contradicts the abstract's claim of 'superior or comparable' performance relative to classical ML algorithms.
  3. [Section III.A and III.B] No statistical uncertainty is reported for any of the quantum results. With a dataset of only 130 samples, single 3-fold cross-validation runs can have high variance, and the precision/recall/F1 ranges (e.g., recall from 0.25 to 1 for the full-feature model) indicate that some classes are poorly recognized. The paper does not provide standard deviations, confidence intervals, or repeated-run results, so the reported differences between settings (e.g., 0.79 vs. 0.86 validation accuracy) and the comparison with classical baselines are not statistically grounded.
minor comments (4)
  1. [Section II.A.2 / Conclusion] The number of PCA components is not stated in the preprocessing section; it first appears as '65 dimensions' in the conclusion. The methods should specify how many components were retained when keeping 95% of the variance.
  2. [Eq. (3)] The denominator in Eq. (3) is difficult to read: the summation notation appears as 'P5' rather than a proper sum over k. Please clarify the formula.
  3. [Section II.B.2 and Fig. 2] The paper describes two hardware-efficient ansatze with different rotation gates, but Fig. 2 only shows the first layer and does not clearly depict both HEA structures or the mapping of the 15 qubits to the readout of five classes. A complete circuit diagram for both ansatze would improve reproducibility.
  4. [Abstract / Conclusion] The phrase 'quantum-enhanced' is not justified by any comparison showing a quantum advantage over classical methods; the reported results are mostly comparable or worse than the classical baselines listed in Table I. The language should be softened to reflect a feasibility study.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the results are an empirical pipeline with a separate reproducibility concern, not a self-referential derivation.

full rationale

Walk-through: the paper's chain is empirical (preprocess, amplitude encode, two HEA circuits, softmax readout, gradient-descent training, held-out validation), and none of these steps defines the reported accuracy in terms of the fitted parameters. The validation accuracies of 0.79 and 0.86 come from held-out data, so they are out-of-sample measurements rather than renamed training fits. The CuMiDa classical baselines are external published results, not outputs of the present model, so the comparison is not circular by construction. Refs [13] and [15] are author self-citations, but they appear only as contextual examples of quantum-AI applications in the introduction; the architecture, experiments, and conclusions do not rest on those works. The amplitude-encoding dimension mismatch (Eq. (1) requires 2^15 = 32,768 components for 15 qubits, while the full-feature pipeline claims 54,676 features) is a serious correctness and reproducibility concern, but it is not circularity: an inconsistent pipeline can be wrong without being self-referential. No equation in the paper forces the reported accuracy to equal an input by construction, and no load-bearing claim reduces to a self-citation chain. Therefore the appropriate finding is no significant circularity.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The ledger reflects that the paper's central performance claim rests on dataset quality, simulator correctness, and an internally inconsistent amplitude encoding assumption. There are no new postulated physical entities.

free parameters (4)
  • Variational parameters θ in the two HEAs = Learned values after gradient descent
    These are trained to minimize cross-entropy on the labeled data, so the reported accuracy reflects fitting, not a parameter-free prediction.
  • Number of qubits = 15
    Chosen by the authors; not derived from the data, and inconsistent with the claim of encoding 54,676 features via amplitude encoding.
  • Number of circuit layers = 25
    Chosen by hand to increase model capacity; no analysis of sensitivity is provided.
  • PCA variance threshold = 95%
    Hand-selected threshold that determines the reduced feature dimension; the paper notes aggressive reduction may lose information.
assumptions (4)
  • domain assumption The CuMiDa dataset provides accurate class labels and gene expression measurements for brain tumor samples.
    The classification results depend on the integrity of this public dataset, which the paper uses without independent validation.
  • domain assumption The PennyLane state-vector simulator correctly implements the quantum circuit operations.
    All results are generated on the default.qubit simulator, so the correctness of the simulated quantum mechanics is assumed.
  • standard math Amplitude encoding maps normalized classical data into quantum state amplitudes as described by Equation (1).
    This is a standard quantum information encoding, though applied inconsistently to a 15-qubit system with 54,676 features.
  • ad hoc to paper The published classical ML baselines in the CuMiDa database are directly comparable to the quantum model's cross-validation results.
    The paper compares accuracy numbers from different sources without showing that train/validation splits and preprocessing match.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum-Enhanced Classification of Brain Tumors Using DNA Microarray Gene Expression Profiles." pith.science (2026). https://pith.science/paper/LAGVK7C3

@misc{pith2026250502033,
  author       = {Pith},
  title        = {Pith review of: Quantum-Enhanced Classification of Brain Tumors Using DNA Microarray Gene Expression Profiles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LAGVK7C3}},
  note         = {Machine review of arXiv:2505.02033}
}
read the original abstract

DNA microarray technology enables the simultaneous measurement of expression levels of thousands of genes, thereby facilitating the understanding of the molecular mechanisms underlying complex diseases such as brain tumors and the identification of diagnostic genetic signatures. To derive meaningful biological insights from the high-dimensional and complex gene features obtained through this technology and to analyze gene properties in detail, classical AI-based approaches such as machine learning and deep learning are widely employed. However, these methods face various limitations in managing high-dimensional vector spaces and modeling the intricate relationships among genes. In particular, challenges such as hyperparameter tuning, computational costs, and high processing power requirements can hinder their efficiency. To overcome these limitations, quantum computing and quantum AI approaches are gaining increasing attention. Leveraging quantum properties such as superposition and entanglement, quantum methods enable more efficient parallel processing of high-dimensional data and offer faster and more effective solutions to problems that are computationally demanding for classical methods. In this study, a novel model called "Deep VQC" is proposed, based on the Variational Quantum Classifier approach. Developed using microarray data containing 54,676 gene features, the model successfully classified four different types of brain tumors-ependymoma, glioblastoma, medulloblastoma, and pilocytic astrocytoma-alongside healthy samples with high accuracy. Furthermore, compared to classical ML algorithms, our model demonstrated either superior or comparable classification performance. These results highlight the potential of quantum AI methods as an effective and promising approach for the analysis and classification of complex structures such as brain tumors based on gene expression features.

Figures

Figures reproduced from arXiv: 2505.02033 by the authors.

Figure 1
Figure 1. Class distribution of the microarray dataset, which includes samples [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Representation of the proposed Deep VQC model. Due to space constraints, only the first layer is shown. The complete model consists of 25 layers. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Training accuracy and cost curves of the Deep VQC model. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Validation accuracy and cost curves of the Deep VQC model. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Confusion matrix of the Deep VQC model. algorithms from the CuMiDa database. The accuracy values for these algorithms are presented in Table I. IV. CONCLUSION In this study, a deep variational quantum classification model—referred to as Deep VQC—was developed based on …
Figure 6
Figure 6. Figure 6: Training accuracy and cost curves of the Deep VQC Model with PCA. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Validation accuracy and cost curves of the Deep VQC Model with [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 18 canonical work pages

  1. [1]

    Whole-genome expression microarray combined with machine learning to identify prognostic biomarkers for high-grade glioma,

    C. Shu, Q. Wang, X. Yan, and J. Wang, “Whole-genome expression microarray combined with machine learning to identify prognostic biomarkers for high-grade glioma,” Journal of Molecular Neuroscience, vol. 64, no. 4, pp. 491–500, Apr 2018

  2. [2]

    Machine learning in dna microarray anal- ysis for cancer classification,

    S.-B. Cho and H.-H. Won, “Machine learning in dna microarray anal- ysis for cancer classification,” in Proceedings of the First Asia-Pacific Bioinformatics Conference on Bioinformatics 2003-Volume 19, 2003, pp. 189–198

  3. [3]

    New developments in microarray technology,

    D. Blohm and A. Guiseppi-Elie, “New developments in microarray technology,” Current Opinion in Biotechnology , vol. 12, no. 1, pp. 41– 47, Feb 2001

  4. [4]

    Detection of genetic and chromosomal aber- rations in medulloblastomas and primitive neuroectodermal tumors with dna microarrays,

    N. Kagawa, M. Maruno, T. Suzuki, T. Hashiba, N. Hashimoto, S. Izu- moto, and T. Yoshimine, “Detection of genetic and chromosomal aber- rations in medulloblastomas and primitive neuroectodermal tumors with dna microarrays,” Brain Tumor Pathology , vol. 23, no. 1, pp. 41–47, 2006

  5. [5]

    A top-r feature selection algo- rithm for microarray gene expression data,

    A. Sharma, S. Imoto, and S. Miyano, “A top-r feature selection algo- rithm for microarray gene expression data,” IEEE-ACM Transactions on Computational Biology and Bioinformatics , vol. 9, no. 3, pp. 754–764, May-Jun 2012

  6. [6]

    Deep learning techniques for cancer classification using microarray gene expression data,

    S. Gupta, M. K. Gupta, M. Shabaz, and A. Sharma, “Deep learning techniques for cancer classification using microarray gene expression data,” Frontiers in Physiology, vol. 13, Sep 2022

  7. [7]

    Gene selection and classification for cancer microarray data based on machine learning and similarity measures,

    Q. Liu, A. H. Sung, Z. Chen, J. Liu, L. Chen, M. Qiao, Z. Wang, X. Huang, and Y . Deng, “Gene selection and classification for cancer microarray data based on machine learning and similarity measures,” BMC Genomics, vol. 12, no. 5, Dec 2011

  8. [8]

    Machine learning model for cancer diagnosis based on rnaseq microarray,

    H. Torkey, M. Atlam, N. El-Fishawy, and H. Salem, “Machine learning model for cancer diagnosis based on rnaseq microarray,” Menoufia Journal of Electronic Engineering Research , vol. 30, no. 1, pp. 65–75,

Show all 21 references
  1. [9]

    Deep learning approach for brain tumor classification using metaheuristic optimization with gene expression data,

    A. A. Joshi and R. M. Aziz, “Deep learning approach for brain tumor classification using metaheuristic optimization with gene expression data,” International Journal of Imaging Systems and Technology, vol. 34, no. 2, Mar 2024

  2. [10]

    Integrating multiple microarray dataset analysis and machine learning methods to reveal the key genes and regula- tory mechanisms underlying human intervertebral disc degeneration,

    H. Chang, X. Yang, K. You, M. Jiang, F. Cai, Y . Zhang, L. Liu, H. Liu, and X. Liu, “Integrating multiple microarray dataset analysis and machine learning methods to reveal the key genes and regula- tory mechanisms underlying human intervertebral disc degeneration,” PEERJ, vol...

  3. [11]

    Quantum computing: A taxonomy, systematic review and future directions,

    S. S. Gill, A. Kumar, H. Singh, M. Singh, K. Kaur, M. Usman, and R. Buyya, “Quantum computing: A taxonomy, systematic review and future directions,” Software-Practice & Experience , vol. 52, no. 1, pp. 66–114, Jan 2022

  4. [12]

    Multi-omic and quantum machine learning integration for lung subtypes classification,

    M. K. Saggi, A. S. Bhatia, M. Isaiah, H. Gowher, and S. Kais, “Multi-omic and quantum machine learning integration for lung subtypes classification,” 2024. [Online]. Available: https://arxiv.org/abs/ 2410.02085

  5. [13]

    Beyond limits: Charting new horizons in glioma tumor classification through hybrid quantum computing with The Cancer Genome Atlas (TCGA) data,

    E. Akpinar and M. Oduncuoglu, “Beyond limits: Charting new horizons in glioma tumor classification through hybrid quantum computing with The Cancer Genome Atlas (TCGA) data,” In Review, Jan 2024

  6. [14]

    Hqnet: A hybrid quantum network for multi-class mri brain classification via quantum computing,

    A. Wang, D. Mao, X. Li, T. Li, and L. Li, “Hqnet: A hybrid quantum network for multi-class mri brain classification via quantum computing,” Expert Systems with Applications , vol. 261, FEB 1 2025

  7. [15]

    The role of quantum- enhanced support vector machine using multiparametric mri parameters in differentiating medulloblastoma from ependymoma,

    E. Akpinar, N. M. Duc, and B. Keserci, “The role of quantum- enhanced support vector machine using multiparametric mri parameters in differentiating medulloblastoma from ependymoma,” in 2022 IEEE International Conference on Quantum Computing and Engineering (QCE 2022), 2022, p...

  8. [16]

    Cumida: An extensively curated microarray database for benchmarking and testing of machine learning approaches in cancer research,

    B. C. Feltes, E. B. Chandelier, B. I. Grisci, and M. Dorn, “Cumida: An extensively curated microarray database for benchmarking and testing of machine learning approaches in cancer research,” Journal of Computational Biology, vol. 26, no. 4, pp. 376–386, Apr 2019

  9. [17]

    Variational quantum algorithms,

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, and P. J. Coles, “Variational quantum algorithms,” Nature Reviews Physics, vol. 3, no. 9, pp. 625–644, Sep 2021

  10. [18]

    Schuld and F

    M. Schuld and F. Petruccione, Machine learning with quantum comput- ers. Springer, 2021, vol. 676

  11. [19]

    Noisy intermediate-scale quantum algorithms,

    K. Bharti, A. Cervera-Lierta, T. H. Kyaw, T. Haug, S. Alperin- Lea, A. Anand, M. Degroote, H. Heimonen, J. S. Kottmann, T. Menke, W.-K. Mok, S. Sim, L.-C. Kwek, and A. Aspuru- Guzik, “Noisy intermediate-scale quantum algorithms,” Rev. Mod. Phys., vol. 94, p. 015004, Feb 2022. ...

  12. [20]

    Quantum chemistry in the age of quantum computing,

    Y . Cao, J. Romero, J. P. Olson, M. Degroote, P. D. Johnson, M. Kiefer- ova, I. D. Kivlichan, T. Menke, B. Peropadre, N. P. D. Sawaya, S. Sim, L. Veis, and A. Aspuru-Guzik, “Quantum chemistry in the age of quantum computing,” Chemical Reviews, vol. 119, no. 19, pp. 10 856– 10 ...

  13. [2021]

    Available: https://mjeer.journals.ekb.eg/article 146277

    [Online]. Available: https://mjeer.journals.ekb.eg/article 146277. html

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.