Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Knowledge-Guided Biomarker Identification for Label-Free Single-Cell RNA-Seq Data: A Reinforcement Learning Perspective

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper proposes RIGPS, a label-free gene panel selection method that combines ensembled prior knowledge from existing selectors with multi-agent reinforcement learning, and claims it outperforms all baselines on 24 single-cell RNA-seq…

desk verdict A genuinely new RL-based gene panel selector, but Algorithm 1's unspecified final-selection metric makes the reported gains unverifiable. read the letter →

arxiv 2501.04718 v2 pith:BAXDIX5T submitted 2025-01-02 q-bio.GN cs.AI

Meng Xiao , Weiliang Zhang , Xiaohan Huang , Hengshu Zhu , Min Wu , Xiaoli Li , Yuanchun Zhou This is my paper · ORCID
classification q-bio.GNcs.AI
keywords genepanelselectionsingle-cellRNA-seqreinforcementlearningmulti-agentlabel-freebiomarkerdiscoveryclusteringknowledgeensemblescfeature
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

RIGPS is an iterative, label-free gene panel selection method for single-cell RNA-seq. It first combines several existing gene-selection algorithms into a coarse candidate gene set, then uses multiple reinforcement-learning agents to refine that set through repeated clustering and reward feedback. The paper claims that on 24 public datasets the resulting panels outperform all tested baselines on clustering quality, rank first in NMI on 19 of 24 datasets, and also improve supervised cell-type annotation, while producing smaller gene panels. If correct, this would give biologists compact marker sets for cheaper targeted assays and more robust downstream analysis without needing cell-type labels.

What carries the argument

The load-bearing mechanism is a multi-agent reinforcement-learning loop in which one agent per candidate gene decides to select or discard its gene, sharing a state representation built from descriptive statistics of the current gene subset and compressed by an autoencoder. The reward combines a biological-distinctiveness term, the normalized mutual information between the selected genes' expression and Louvain pseudo-labels, and a parsimony term that penalizes large panels; the loop is bootstrapped by injecting gene subsets produced by existing selectors as initial experiences. That combination lets the search start from ensemble knowledge while still moving beyond it.

What would settle it

Rerun RIGPS with the final selection step restricted to rewards computed only from pseudo-labels (no true labels anywhere), and compare the resulting NMI on the same 24 datasets; if the margin over the baselines collapses, the reported advantage depends on label-informed selection rather than on the reinforcement-learning pipeline.

Watch

Extended reading notes

Core claim

The central claim is that RIGPS finds a better gene panel than existing methods precisely because it treats panel selection as an expert-like sequential decision process: ensemble knowledge from simpler selectors provides a high-recall starting boundary, multi-agent reinforcement learning then explores and optimizes within that boundary, and a reward function that combines an unsupervised separability term with a parsimony term guides the agents toward genes that distinguish cell populations. The paper reports that this pipeline achieves the highest average NMI, ARI, and silhouette scores across 24 scRNA-seq datasets from multiple species and tissues, with the best NMI rank on 19 of 24 datasets, and that it surpasses all baselines on balanced accuracy and F1 metrics for cell-type annotation.

Load-bearing premise

Algorithm 1's final step picks the candidate gene panel with the best clustering metric, but the paper never says whether that metric uses the true cell-type labels; if it does, the method is not fully label-free and the benchmark gains could come from that label access.

Editorial extensions

If this is right

  • The selected panels are compact enough to lower the cost of targeted assays, since the paper shows RIGPS chooses fewer genes than the second-best baseline while keeping higher clustering performance.
  • Because the method needs no cell-type labels for training, it can be applied to new tissues and species without manual annotation, a scenario the 24-dataset evaluation is meant to support.
  • Ablation results show that both the pre-filtering step and the knowledge injection contribute to the final gain, implying the ensemble of basic selectors is doing real work rather than being a formality.
  • Reported linear scaling of training time with cell number suggests the approach remains usable on atlas-scale datasets, at least within the sizes tested.

Reading between the lines

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

  • Editorial inference: if the final selection step in Algorithm 1 uses true cell-type labels to pick among candidate panels, the headline 'label-free' claim would be undercut, and the reported advantage might partly reflect that label access rather than the RL refinement.
  • Editorial inference: because the reward is a cluster-quality score computed from pseudo-labels, the method may favor genes that produce the easiest clusters rather than the most biologically meaningful ones, so independent validation against known marker genes or perturbation data would strengthen the biological claim.
  • Editorial inference: the same architecture could be transferred to other high-dimensional molecular measurements, such as spatial transcriptomics or ATAC-seq, by replacing the state statistics while keeping the reward and agent loop unchanged.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes RIGPS, a gene panel selection method for label-free single-cell RNA-seq data that combines an ensemble pre-filtering step, injection of prior knowledge from existing gene selection algorithms, and multi-agent reinforcement learning. Each gene is controlled by an agent that selects or discards it; the reward combines a pseudo-label-based NMI term (Eq. 2) and a parsimony term (Eq. 3). The final panel is chosen in Algorithm 1 by evaluating a clustering metric E over all candidate subsets encountered during training and keeping the best. The authors report that RIGPS outperforms eight baseline methods on 24 scRNA-seq datasets for clustering (NMI, ARI, SI) and for supervised cell-type annotation, alongside ablations, hyperparameter studies, robustness analysis, and biological case studies.

Significance. If the label-free claim and the reported performance hold, RIGPS would be a practically useful contribution: it targets a real problem, is evaluated on 24 datasets across species and tissues, includes ablations of each component, and provides qualitative downstream analyses. The code and selected gene sets are promised to be publicly available, which is a strength for reproducibility. However, the central quantitative claim is currently undermined by an unspecified and potentially label-aware final selection step, and by the alignment between the training reward and the evaluation metric. These issues are load-bearing because the benchmark numbers, rather than the architecture alone, are the main evidence for the method.

major comments (3)
  1. [Algorithm 1, lines 15-21; Section 4.2; Eq. (1)] The final selection step evaluates E(C(X[G'])) over every candidate subset in the history H and returns the subset with the highest E. The paper never states what E is in this step. Section 4.2 defines the evaluation metrics as NMI, ARI, and SI, with NMI and ARI computed against true cell-type labels, while Eq. (1) uses the same E and C notation as Algorithm 1. If E is NMI or ARI, the final gene panel is selected using oracle access to the ground-truth labels that also define the reported performance, which contradicts the label-free claim (Section 3.2, Algorithm 1 input) and makes the comparison against baselines unfair, because the baselines do not receive an equivalent label-informed final selection. If E is instead SI or a pseudo-label-based metric, this must be stated explicitly and the effect of the best-over-history selection on the reported margins must be quantified. As written, both readings are plausible, and this ambiguity is load-bearing for the Section 5.1 claim that RIGPS outperforms all baselines on 19 of 24 datasets in NMI.
  2. [Section 3.2, Eq. (2); Section 4.2; Section 5.1] The reward function is NMI between the selected gene expression matrix and pseudo-labels obtained from Louvain clustering of that same expression matrix, and the benchmark metrics are NMI and ARI computed from Louvain clustering of the selected genes. This means the agent is trained to maximize essentially the same quantity on which it is evaluated. Combined with the best-over-history selection in Algorithm 1, the reported gains may be partly a selection artifact rather than evidence of independent predictive quality. Please report the performance of the final trained policy without best-over-history selection, or using a held-out evaluation protocol, so that the reader can separate the contribution of the learned policy from the contribution of the selection rule.
  3. [Section 5.1 and Table 7] All clustering and annotation results are reported as single point estimates. No standard deviations, confidence intervals, or numbers of independent RL runs are given. Since the policies are stochastic and the final panel is selected by scanning the training history, the claim that the average NMI of RIGPS outperforms all baselines needs variance quantification to establish that the difference is not within run-to-run noise. This is particularly important because the Nemenyi test in Figure 3 is presumably computed from a single set of per-dataset values.
minor comments (5)
  1. [Section 5.1 and Figure 3 caption] The text says Figure 3 compares RIGPS with ten gene panel selection methods, but Section 4.3 lists eight baselines and Table 7 additionally reports Original Dataset and Random, so the count is inconsistent.
  2. [Section 3.3, Eq. (5)] The denominator in Eq. (5) is written as 'P rj ∈P rj', which is confusing; it should be the sum of r_j over all j in P, written as Σ_j r_j.
  3. [Section 4.4] The sentence 'We adopt the Louvain community detection algorithm to generate pseudo-labels for reward estimation and those supervised selection methods' should clarify whether the Random Forest, SVM, RFE, and KBest pre-filtering methods are trained on Louvain pseudo-labels from the full dataset or from a candidate subset. This is important for assessing possible information leakage in the pre-filtering step.
  4. [Section 3.3 and Eq. (7)] The 2-sigma threshold in Eq. (7) is described as forming a coarse boundary with high recall, but no justification or empirical support is given for the 2-sigma choice; the hyperparameter study in Section 5.8 varies the set of basic methods but not the threshold.
  5. [References] Reference [46] is listed with a CrossRef 'Deleted DOI' placeholder; a proper citation for normalized mutual information should be provided.

Circularity Check

1 steps flagged · score 7.0 of 10

Algorithm 1's final selection maximizes the same unspecified metric E used for evaluation; the reported NMI/ARI gains are partly selection results rather than independent predictions.

  1. fitted input called prediction [Algorithm 1, lines 15-21; Eq. (1); Section 4.2]
    "G∗ = argmax G′⊆G E(C(X[G ′])) [Eq. 1] ... foreach G ′ ∈ H do Evaluate clustering metric E(C(X[G ′])); if E > P ∗ then P ∗ ← E; G ∗ ← G ′ [Algorithm 1, lines 16-20]"

    The paper's headline claim ('RIGPS achieves the highest rank on 19 out of 24 datasets... in terms of NMI') is measured with NMI/ARI computed against true cell-type labels (Section 4.2). Yet Eq. (1) defines the selection problem as maximizing the same E, and Algorithm 1 returns the explored subset with maximum E. If E is NMI or ARI, then the reported NMI/ARI is, by construction, the best value of the evaluation metric over the history, not an independent prediction of panel quality. If E is instead a label-free metric such as SI, the paper never states this and never quantifies the optimism from best-over-history selection. Either way, the central benchmark numbers are partly produced by the selection rule itself, and the label-free claim is left ambiguous.

full rationale

The core RL loop (policy-gradient training, pseudo-labeled NMI reward, knowledge injection) is not circular by itself: the reward in Eq. (2) uses Louvain pseudo-labels rather than true labels, and the policy is trained before the final evaluation. The circularity burden sits on the final selection in Algorithm 1. Eq. (1) defines the gene-panel problem as G* = argmax E(C(X[G'])), and Algorithm 1 implements this by evaluating E over history H and returning the best subset. Section 4.2 states that the adopted evaluation metrics are NMI, ARI, and SI, with NMI and ARI computed against true cell-type labels. The paper never specifies which E is used in Algorithm 1. If E is NMI or ARI, then the reported 'RIGPS achieves the highest rank on 19 out of 24 datasets in terms of NMI' is the maximum of the evaluation metric over explored subsets, so the central benchmark claim reduces by construction to the selection rule: the method chooses the panel that already scores best on the reported yardstick. If E is SI or a pseudo-label-based NMI, the final step is an optimistic best-over-history selection whose effect on the reported margins is never quantified. Either reading makes the benchmark evidence partially circular. The self-citations [72] and [73] are used only as related-work inspiration and are not load-bearing, so they do not increase the score. I assign 7 rather than 8 because the RL exploration does genuinely search a subset space, and the reduction is conditional on the unspecified choice of E.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the validity of Louvain pseudo-labels as a proxy for cell types, the sufficiency of the ensemble pre-filter, and the compression of gene states into descriptive statistics. No new biological entities are introduced. The final selection step in Algorithm 1 is the most consequential design choice because it may use the evaluation labels themselves.

free parameters (4)
  • alpha (reward balance) = 0.5
    Chosen via hyperparameter study on Cao dataset (Section 5.6) and fixed across all datasets.
  • lambda (parsimony penalty) = 0.7
    Reported in Section 5.6 after the study on Cao; not listed in the main hyperparameter section 4.4.
  • 2-sigma meta-vote threshold = 2
    Coarse boundary keeps genes with meta-vote scores above mu plus 2 sigma (Section 3.3); this fixed threshold is hand-chosen.
  • state latent dimension k = unspecified
    Autoencoder latent size k is described as fixed but its value is never given (Section 3.1).
assumptions (5)
  • domain assumption Louvain clustering on the selected gene panel produces meaningful pseudo-labels that correspond to real cell types.
    The reward (Eq. 2) and knowledge weights (Section 3.3) depend on pseudo-labels from Louvain; if these labels are not biologically meaningful, the reward is misaligned.
  • domain assumption NMI between gene expression and pseudo-labels is a valid measure of biological distinctiveness.
    The reward function replaces expert judgment with this NMI quantity, but no independent validation of the surrogate is given.
  • domain assumption The ensemble of Random Forest, SVM, RFE, geneBasis, and KBest, combined with a 2-sigma cutoff, retains the informative genes with high recall.
    Section 3.3 asserts this without measuring recall against any reference set.
  • domain assumption Descriptive statistics of the selected gene subset (std, min, max, quartiles) preserve enough information for gene selection.
    The state representation in Section 3.1 discards raw expression; the paper does not show that this compression is lossless enough for the task.
  • domain assumption Ground-truth cell type labels used for evaluation are correct and complete.
    All benchmark metrics in Tables 7 and 1 compare against these labels; label noise would change all scores.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Knowledge-Guided Biomarker Identification for Label-Free Single-Cell RNA-Seq Data: A Reinforcement Learning Perspective." pith.science (2026). https://pith.science/paper/BAXDIX5T

@misc{pith2026250104718,
  author       = {Pith},
  title        = {Pith review of: Knowledge-Guided Biomarker Identification for Label-Free Single-Cell RNA-Seq Data: A Reinforcement Learning Perspective},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BAXDIX5T}},
  note         = {Machine review of arXiv:2501.04718}
}
read the original abstract

Gene panel selection aims to identify the most informative genomic biomarkers in label-free genomic datasets. Traditional approaches, which rely on domain expertise, embedded machine learning models, or heuristic-based iterative optimization, often introduce biases and inefficiencies, potentially obscuring critical biological signals. To address these challenges, we present an iterative gene panel selection strategy that harnesses ensemble knowledge from existing gene selection algorithms to establish preliminary boundaries or prior knowledge, which guide the initial search space. Subsequently, we incorporate reinforcement learning through a reward function shaped by expert behavior, enabling dynamic refinement and targeted selection of gene panels. This integration mitigates biases stemming from initial boundaries while capitalizing on RL's stochastic adaptability. Comprehensive comparative experiments, case studies, and downstream analyses demonstrate the effectiveness of our method, highlighting its improved precision and efficiency for label-free biomarker discovery. Our results underscore the potential of this approach to advance single-cell genomics data analysis.

Figures

Figures reproduced from arXiv: 2501.04718 by the authors.

Figure 1
Figure 1. The Motivation of our study is the imitation of expert [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The overview of our framework. (a) After raw data preprocessing, we filter the irrelevant genes and collect expert [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overall quantitative analysis on downstream clustering task performance comparison: (a-c) Comparison of [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: Ablation studies of RIGPS in terms of NMI. The comparison between RIGPS−a , RIGPS−r , and RIGPS shows the impact of reinforced optimization. The comparison between RIGPS−k and RIGPS shows the impact of the knowledge injection. The comparison between RIGPS−f and RIGPS s…
Figure 5
Figure 5. Figure 5: Model robustness checks on the Human Pancreas dataset under batch effect circumstances. (a) The visualization of [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Iterative convergence speed comparison between [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: The comparison of the selected result in the coarse [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: The result of the hyperparameter sensitivity test on [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Scalability check of RIGPS regarding parameter size and training time. 5 10 70 75 80 85 90 95 100 NMI ARI SI RIGPS RIGPS b RIGPS a (a) Leng 0 20 25 30 35 40 45 50 NMI ARI SI RIGPS RIGPS b RIGPS a (b) Maria2 15 20 65 70 75 80 85 90 NMI ARI SI RIGPS RIGPS b RIGPS a (c) M…
Figure 10
Figure 10. Figure 10: Performance comparison between three pre-filtering strategies. [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: Comparison with different knowledge injection settings. (a-d) The performance of [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: We found that the gene panel obtained by RIGPS is significantly more compact than the second best while still outperforming it. We speculate the underlying driver for this observation is that gene pre-filtering will remove many redundant genes. Then, our reinforcement…
Figure 13
Figure 13. Figure 13: Differential expression analysis. We highlight the [PITH_FULL_IMAGE:figures/full_fig_p010_13.png]
Figure 14
Figure 14. Figure 14: Visualization analysis of the Puram dataset. (a) t-SNE [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]
Figure 15
Figure 15. Figure 15: Heatmap analysis of the Puram dataset. (a) expres [PITH_FULL_IMAGE:figures/full_fig_p011_15.png]
Figure 16
Figure 16. Figure 16: Expression differential analysis of the rest datasets. [PITH_FULL_IMAGE:figures/full_fig_p024_16.png]
Figure 17
Figure 17. Figure 17: t-SNE visualization of the rest datasets, where the figure in the left panel is visualized from the original dataset. [PITH_FULL_IMAGE:figures/full_fig_p025_17.png]
Figure 18
Figure 18. Figure 18: (1/2) Expression heatmap of the rest datasets, where the figure in the left panel is visualized from the original [PITH_FULL_IMAGE:figures/full_fig_p026_18.png]
Figure 19
Figure 19. Figure 19: (2/2) Expression heatmap of the rest datasets, where the figure in the left panel is visualized from the original [PITH_FULL_IMAGE:figures/full_fig_p027_19.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Towards Data-Centric AI: A Comprehensive Survey of Traditional, Reinforcement, and Generative Approaches for Tabular Data Transformation

    cs.LG 2025-01 conditional novelty 2.0 of 10

    A survey of feature selection and feature generation for tabular data, categorizing traditional, RL-based, and generative AI methods into a unified taxonomy.

Reference graph

Works this paper leans on

76 extracted references · 69 canonical work pages · cited by 1 Pith paper

  1. [1]

    Single-cell epigenomics: techniques and emerging applications,

    O. Schwartzman and A. Tanay, “Single-cell epigenomics: techniques and emerging applications,”Nature Reviews Genetics, vol. 16, no. 12, pp. 716–726, 2015

    work page 2015

  2. [2]

    Single-cell genome sequenc- ing: current state of the science,

    C. Gawad, W. Koh, and S. R. Quake, “Single-cell genome sequenc- ing: current state of the science,”Nature Reviews Genetics, vol. 17, no. 3, pp. 175–188, 2016

    work page 2016

  3. [3]

    Building a lineage from single cells: genetic techniques for cell lineage tracking,

    M. B. Woodworth, K. M. Girskis, and C. A. Walsh, “Building a lineage from single cells: genetic techniques for cell lineage tracking,”Nature Reviews Genetics, vol. 18, no. 4, pp. 230–244, 2017

    work page 2017

  4. [4]

    Gut microbiota and tuberculosis,

    Y. Liu, L. Yang, M. Meskini, A. Goel, M. Opperman, S. S. Shyamal, A. Manaithiya, M. Xiao, R. Ni, Y. Anet al., “Gut microbiota and tuberculosis,”iMeta, p. e70054

    work page

  5. [5]

    Single-cell multiomics: technologies and data analysis methods,

    J. Lee, D. Y. Hyeon, and D. Hwang, “Single-cell multiomics: technologies and data analysis methods,”Experimental & Molecular Medicine, vol. 52, no. 9, pp. 1428–1442, 2020

    work page 2020

  6. [6]

    The technological landscape and applications of single-cell multi-omics,

    A. Baysoy, Z. Bai, R. Satija, and R. Fan, “The technological landscape and applications of single-cell multi-omics,”Nature Reviews Molecular Cell Biology, vol. 24, no. 10, pp. 695–713, 2023

    work page 2023

  7. [7]

    Topological and geometric analysis of cell states in single-cell transcriptomic data,

    T. Huynh and Z. Cang, “Topological and geometric analysis of cell states in single-cell transcriptomic data,”Briefings in Bioinformatics, vol. 25, no. 3, p. bbae176, 2024

    work page 2024

  8. [8]

    Exploring tissue architecture using spatial transcriptomics,

    A. Rao, D. Barkley, G. S. Fran c ¸a, and I. Yanai, “Exploring tissue architecture using spatial transcriptomics,”Nature, vol. 596, no. 7871, pp. 211–220, 2021

    work page 2021

Show all 76 references
  1. [9]

    Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics,

    S. K. Longo, M. G. Guo, A. L. Ji, and P . A. Khavari, “Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics,”Nature Reviews Genetics, vol. 22, no. 10, pp. 627– 644, 2021

    2021

  2. [10]

    Identifying phenotype-associated subpopulations by integrating bulk and single-cell sequencing data,

    D. Sun, X. Guan, A. E. Moran, L.-Y. Wu, D. Z. Qian, P . Schedin, M.-S. Dai, A. V . Danilov, J. J. Alumkal, A. C. Adeyet al., “Identifying phenotype-associated subpopulations by integrating bulk and single-cell sequencing data,”Nature biotechnology, vol. 40, no. 4, pp. 527–538, 2022

    2022

  3. [11]

    scgpt: Towards building a foundation model for single-cell multi-omics using generative ai,

    H. Cui, C. Wang, H. Maan, K. Pang, F. Luo, and B. Wang, “scgpt: Towards building a foundation model for single-cell multi-omics using generative ai,”bioRxiv, pp. 2023–04, 2023

    2023

  4. [12]

    Transfer learning enables predictions in network biology,

    C. V . Theodoris, L. Xiao, A. Chopra, M. D. Chaffin, Z. R. Al Sayed, M. C. Hill, H. Mantineo, E. M. Brydon, Z. Zeng, X. S. Liuet al., “Transfer learning enables predictions in network biology,”Nature, pp. 1–9, 2023

    2023

  5. [13]

    Genecompass: deciphering universal gene regulatory mechanisms with a knowledge-informed cross-species foundation model,

    X. Yang, G. Liu, G. Feng, D. Bu, P . Wang, J. Jiang, S. Chen, Q. Yang, H. Miao, Y. Zhanget al., “Genecompass: deciphering universal gene regulatory mechanisms with a knowledge-informed cross-species foundation model,”Cell Research, pp. 1–16, 2024

    2024

  6. [14]

    Scihorizon: Benchmarking ai-for- science readiness from scientific data to large language models,

    C. Qin, X. Chen, C. Wang, P . Wu, X. Chen, Y. Cheng, J. Zhao, M. Xiao, X. Dong, Q. Longet al., “Scihorizon: Benchmarking ai-for- science readiness from scientific data to large language models,” inProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mi...

    2025

  7. [15]

    Challenges in unsupervised clustering of single-cell rna-seq data,

    V . Y. Kiselev, T. S. Andrews, and M. Hemberg, “Challenges in unsupervised clustering of single-cell rna-seq data,”Nature Reviews Genetics, vol. 20, no. 5, pp. 273–282, 2019. 13

    2019

  8. [16]

    Sparse pca via l2,p- norm regularization for unsupervised feature selection,

    Z. Li, F. Nie, J. Bian, D. Wu, and X. Li, “Sparse pca via l2,p- norm regularization for unsupervised feature selection,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 4, pp. 5322–5328, 2021

    2021

  9. [17]

    Principal components analysis (pca),

    A. Ma´ckiewicz and W. Ratajczak, “Principal components analysis (pca),”Computers & Geosciences, vol. 19, no. 3, pp. 303–342, 1993

    1993

  10. [18]

    The art of using t-sne for single-cell transcriptomics,

    D. Kobak and P . Berens, “The art of using t-sne for single-cell transcriptomics,”Nature communications, vol. 10, no. 1, p. 5416, 2019

    2019

  11. [19]

    Dimensionality reduction for visualizing single-cell data using umap,

    E. Becht, L. McInnes, J. Healy, C.-A. Dutertre, I. W. Kwok, L. G. Ng, F. Ginhoux, and E. W. Newell, “Dimensionality reduction for visualizing single-cell data using umap,”Nature biotechnology, vol. 37, no. 1, pp. 38–44, 2019

    2019

  12. [20]

    Fold change and p-value cutoffs significantly alter microarray interpretations,

    M. R. Dalman, A. Deeter, G. Nimishakavi, and Z.-H. Duan, “Fold change and p-value cutoffs significantly alter microarray interpretations,” inBMC bioinformatics, vol. 13. Springer, 2012, pp. 1–4

    2012

  13. [21]

    Evaluation of tools for highly variable gene discovery from single-cell rna-seq data,

    S. H. Yip, P . C. Sham, and J. Wang, “Evaluation of tools for highly variable gene discovery from single-cell rna-seq data,”Briefings in bioinformatics, vol. 20, no. 4, pp. 1583–1589, 2019

    2019

  14. [22]

    Comprehensive integration of single-cell data,

    T. Stuart, A. Butler, P . Hoffman, C. Hafemeister, E. Papalexi, W. M. Mauck, Y. Hao, M. Stoeckius, P . Smibert, and R. Satija, “Comprehensive integration of single-cell data,”cell, vol. 177, no. 7, pp. 1888–1902, 2019

    1902

  15. [23]

    Benchmarking atlas-level data integration in single- cell genomics,

    M. D. Luecken, M. B ¨uttner, K. Chaichoompu, A. Danese, M. Inter- landi, M. F. M ¨uller, D. C. Strobl, L. Zappia, M. Dugas, M. Colom ´e- Tatch´eet al., “Benchmarking atlas-level data integration in single- cell genomics,”Nature methods, vol. 19, no. 1, pp. 41–50, 2022

    2022

  16. [24]

    Predicting cellular responses to novel drug perturbations at a single-cell resolution,

    L. Hetzel, S. Boehm, N. Kilbertus, S. G ¨unnemann, F. Theiset al., “Predicting cellular responses to novel drug perturbations at a single-cell resolution,”Advances in Neural Information Processing Systems, vol. 35, pp. 26 711–26 722, 2022

    2022

  17. [25]

    xtrimogene: An efficient and scalable representation learner for single-cell rna-seq data,

    J. Gong, M. Hao, X. Cheng, X. Zeng, C. Liu, J. Ma, X. Zhang, T. Wang, and L. Song, “xtrimogene: An efficient and scalable representation learner for single-cell rna-seq data,”Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  18. [26]

    Cellbrf: a feature selection method for single-cell clustering using cell balance and random forest,

    Y. Xu, H.-D. Li, C.-X. Lin, R. Zheng, Y. Li, J. Xu, and J. Wang, “Cellbrf: a feature selection method for single-cell clustering using cell balance and random forest,” Bioinformatics, p. i368–i376, Jun 2023. [Online]. Available: http://dx.doi.org/10.1093/bioinformatics/btad216

    2023 doi

  19. [27]

    genebasis: an iterative approach for unsupervised selection of targeted gene panels from scrna-seq,

    A. Missarova, J. Jain, A. Butler, S. Ghazanfar, T. Stuart, M. Brusko, C. Wasserfall, H. Nick, T. Brusko, M. Atkinsonet al., “genebasis: an iterative approach for unsupervised selection of targeted gene panels from scrna-seq,”Genome biology, vol. 22, pp. 1–22, 2021

    2021

  20. [28]

    Highly regional genes: graph-based gene selection for single-cell rna-seq data,

    Y. Wu, Q. Hu, S. Wang, C. Liu, Y. Shan, W. Guo, R. Jiang, X. Wang, and J. Gu, “Highly regional genes: graph-based gene selection for single-cell rna-seq data,”Journal of Genetics and Genomics, p. 891–899, Sep 2022. [Online]. Available: http://dx.doi.org/10.1016/j.jgg.2022.01.004

    2022 doi

  21. [29]

    Single-cell signature explorer for comprehensive visualization of single cell signatures across scrna-seq datasets,

    F. Pont, M. Tosolini, and J. J. Fourni´e, “Single-cell signature explorer for comprehensive visualization of single cell signatures across scrna-seq datasets,”Nucleic acids research, vol. 47, no. 21, pp. e133– e133, 2019

    2019

  22. [30]

    Alphaflow: au- tonomous discovery and optimization of multi-step chemistry using a self-driven fluidic lab guided by reinforcement learning,

    A. A. Volk, R. W. Epps, D. T. Yonemoto, B. S. Masters, F. N. Castellano, K. G. Reyes, and M. Abolhasani, “Alphaflow: au- tonomous discovery and optimization of multi-step chemistry using a self-driven fluidic lab guided by reinforcement learning,” Nature Communications, vol. 1...

    2023

  23. [31]

    A survey on deep reinforcement learning for data processing and analytics,

    Q. Cai, C. Cui, Y. Xiong, W. Wang, Z. Xie, and M. Zhang, “A survey on deep reinforcement learning for data processing and analytics,” IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 5, pp. 4446–4465, 2022

    2022

  24. [32]

    Many-objective jaccard- based evolutionary feature selection for high-dimensional imbal- anced data classification,

    H. Saadatmand and M. Akbarzadeh-T, “Many-objective jaccard- based evolutionary feature selection for high-dimensional imbal- anced data classification,”IEEE Transactions on Pattern Analysis & Machine Intelligence, no. 01, pp. 1–16, 2024

    2024

  25. [33]

    sccompass: An integrated cross-species scrna-seq database for ai-ready,

    P . Wang, W. Liu, J. Wang, Y. Liu, P . Li, P . Xu, W. Cui, R. Zhang, Q. Long, Z. Huet al., “sccompass: An integrated cross-species scrna-seq database for ai-ready,”bioRxiv, pp. 2024–11, 2024

    2024

  26. [35]

    Voting from nearest tasks: Meta-vote pruning of pre-trained models for downstream tasks,

    H. Zhao, T. Zhou, G. Long, J. Jiang, and C. Zhang, “Voting from nearest tasks: Meta-vote pruning of pre-trained models for downstream tasks,” inJoint European Conference on Machine Learning and Knowledge Discovery in Databases. Springer, 2023, pp. 52–68

    2023

  27. [36]

    Prioritized experience replay,

    T. Schaul, J. Quan, I. Antonoglou, and D. Silver, “Prioritized experience replay,”arXiv preprint arXiv:1511.05952, 2015

    2015 arXiv

  28. [37]

    Actor-critic algorithms,

    V . Konda and J. Tsitsiklis, “Actor-critic algorithms,”Advances in neural information processing systems, vol. 12, 1999

    1999

  29. [38]

    Actor-critic models and the a3c: The asynchronous advantage actor-critic model,

    M. Sewak and M. Sewak, “Actor-critic models and the a3c: The asynchronous advantage actor-critic model,”Deep reinforcement learning: frontiers of artificial intelligence, pp. 141–152, 2019

    2019

  30. [39]

    Gene expression omnibus: Ncbi gene expression and hybridization array data repository,

    R. Edgar, M. Domrachev, and A. E. Lash, “Gene expression omnibus: Ncbi gene expression and hybridization array data repository,”Nucleic acids research, vol. 30, no. 1, pp. 207–210, 2002

    2002

  31. [40]

    Arrayexpress—a public repository for microarray gene expression data at the ebi,

    A. Brazma, H. Parkinson, U. Sarkans, M. Shojatalab, J. Vilo, N. Abeygunawardena, E. Holloway, M. Kapushesky, P . Kemmeren, G. G. Laraet al., “Arrayexpress—a public repository for microarray gene expression data at the ebi,”Nucleic acids research, vol. 31, no. 1, pp. 68–71, 2003

    2003

  32. [41]

    The sequence read archive,

    R. Leinonen, H. Sugawara, M. Shumway, and I. N. S. D. Collabora- tion, “The sequence read archive,”Nucleic acids research, vol. 39, no. suppl 1, pp. D19–D21, 2010

    2010

  33. [42]

    Comprehensive single-cell transcriptional profiling of a multicellular organism,

    J. Cao, J. S. Packer, V . Ramani, D. A. Cusanovich, C. Huynh, R. Daza, X. Qiu, C. Lee, S. N. Furlan, F. J. Steemers, A. Adey, R. H. Waterston, C. Trapnell, and J. Shendure, “Comprehensive single-cell transcriptional profiling of a multicellular organism,” Science, p. 661–667, ...

    2017 doi

  34. [43]

    Mapping the mouse cell atlas by microwell-seq,

    X. Han, R. Wang, Y. Zhou, L. Fei, H. Sun, S. Lai, A. Saadatpour, Z. Zhou, H. Chen, F. Ye, D. Huang, Y. Xu, W. Huang, M. Jiang, X. Jiang, J. Mao, Y. Chen, C. Lu, J. Xie, Q. Fang, Y. Wang, R. Yue, T. Li, H. Huang, S. H. Orkin, G.-C. Yuan, M. Chen, and G. Guo, “Mapping the mouse ...

    2018 doi

  35. [44]

    scgen predicts single-cell perturbation responses,

    M. Lotfollahi, F. A. Wolf, and F. J. Theis, “scgen predicts single-cell perturbation responses,”Nature methods, vol. 16, no. 8, pp. 715–721, 2019

    2019

  36. [45]

    Spatial reconstruction of single-cell gene expression data,

    R. Satija, J. Farrell, D. Gennert, A. Schier, and A. Regev, “Spatial reconstruction of single-cell gene expression data,”PMC, Apr 2015

    2015

  37. [46]

    10.1162/153244303321897735,

    A. Strehl and J. Ghosh, “10.1162/153244303321897735,”CrossRef Listing of Deleted DOIs, Jan 2000. [Online]. Available: http: //dx.doi.org/10.1162/153244303321897735

    2000 doi

  38. [47]

    Objective criteria for the evaluation of clustering methods,

    W. M. Rand, “Objective criteria for the evaluation of clustering methods,”Journal of the American Statistical Association, p. 846–850, Dec 1971. [Online]. Available: http://dx.doi.org/10.1080/01621459. 1971.10482356

    1971

  39. [48]

    Silhouettes: A graphical aid to the interpretation and validation of cluster analysis,

    P . J. Rousseeuw, “Silhouettes: A graphical aid to the interpretation and validation of cluster analysis,”Journal of Computational and Applied Mathematics, p. 53–65, Nov 1987. [Online]. Available: http://dx.doi.org/10.1016/0377-0427(87)90125-7

    1987 doi

  40. [49]

    Massively parallel digital transcriptional profiling of single cells,

    G. X. Zheng, J. M. Terry, P . Belgrader, P . Ryvkin, Z. W. Bent, R. Wilson, S. B. Ziraldo, T. D. Wheeler, G. P . McDermott, J. Zhu et al., “Massively parallel digital transcriptional profiling of single cells,”Nature communications, vol. 8, no. 1, p. 14049, 2017

    2017

  41. [50]

    Analytic pearson residuals for normalization of single-cell rna-seq umi data,

    J. Lause, P . Berens, and D. Kobak, “Analytic pearson residuals for normalization of single-cell rna-seq umi data,”Genome biology, vol. 22, pp. 1–20, 2021

    2021

  42. [51]

    Gene panel selection for targeted spatial transcriptomics,

    Y. Zhang, V . Petukhov, E. Biederstedt, R. Que, K. Zhang, and P . V . Kharchenko, “Gene panel selection for targeted spatial transcriptomics,”Genome Biology, vol. 25, no. 1, p. 35, 2024

    2024

  43. [52]

    scgist: gene panel design for spatial transcriptomics with prioritized gene sets,

    M. A. Yafi, M. H. H. Hisham, F. Grisanti, J. F. Martin, A. Rahman, and M. A. H. Samee, “scgist: gene panel design for spatial transcriptomics with prioritized gene sets,”Genome Biology, vol. 25, no. 1, p. 57, 2024

    2024

  44. [53]

    Random forests,

    L. Breiman, “Random forests,”Machine learning, vol. 45, pp. 5–32, 2001

    2001

  45. [54]

    Support vector machine,

    D. A. Pisner and D. M. Schnyer, “Support vector machine,”Machine learning, pp. 101–121, 2020

    2020

  46. [55]

    Recursive feature elimination with random forest for ptr-ms analysis of agroindustrial products,

    P . M. Granitto, C. Furlanello, F. Biasioli, and F. Gasperi, “Recursive feature elimination with random forest for ptr-ms analysis of agroindustrial products,”Chemometrics and intelligent laboratory systems, vol. 83, no. 2, pp. 83–90, 2006

    2006

  47. [56]

    A comparative study on feature selection in text categorization,

    Y. Yang and J. O. Pedersen, “A comparative study on feature selection in text categorization,” inIcml, vol. 97, no. 412-420. Nashville, TN, USA, 1997, p. 35

    1997

  48. [57]

    M3drop: dropout-based feature selection for scrnaseq,

    T. S. Andrews and M. Hemberg, “M3drop: dropout-based feature selection for scrnaseq,”Bioinformatics, p. 2865–2867, Aug 2019. [Online]. Available: http://dx.doi.org/10.1093/bioinformatics/ bty1044 14

    2019 doi

  49. [58]

    Feats: feature selection-based clustering of single-cell rna-seq data,

    E. Vans, A. Patil, and A. Sharma, “Feats: feature selection-based clustering of single-cell rna-seq data,”Briefings in Bioinformatics, Oct

  50. [59]

    Accurate feature selection improves single-cell rna-seq cell clustering,

    K. Su, T. Yu, and H. Wu, “Accurate feature selection improves single-cell rna-seq cell clustering,”Briefings in Bioinformatics, Sep

  51. [60]

    A machine learning method for the discovery of minimum marker gene combinations for cell type identification from single-cell rna sequencing,

    B. Aevermann, Y. Zhang, M. Novotny, M. Keshk, T. Bakken, J. Miller, R. Hodge, B. Lelieveldt, E. Lein, and R. H. Scheuermann, “A machine learning method for the discovery of minimum marker gene combinations for cell type identification from single-cell rna sequencing,”Genome Re...

    2021 doi

  52. [61]

    Reinforcement learning: An introduction,

    R. S. Sutton, “Reinforcement learning: An introduction,”A Bradford Book, 2018

    2018

  53. [62]

    A review of safe reinforcement learning: Methods, theories and applications,

    S. Gu, L. Yang, Y. Du, G. Chen, F. Walter, J. Wang, and A. Knoll, “A review of safe reinforcement learning: Methods, theories and applications,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

  54. [63]

    Structured cooperative reinforcement learning with time-varying composite action space,

    W. Li, X. Wang, B. Jin, D. Luo, and H. Zha, “Structured cooperative reinforcement learning with time-varying composite action space,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 11, pp. 8618–8634, 2021

    2021

  55. [64]

    Fear-neuro-inspired reinforcement learning for safe autonomous driving,

    X. He, J. Wu, Z. Huang, Z. Hu, J. Wang, A. Sangiovanni-Vincentelli, and C. Lv, “Fear-neuro-inspired reinforcement learning for safe autonomous driving,”IEEE transactions on pattern analysis and machine intelligence, 2023

    2023

  56. [65]

    Efficient hierarchical stor- age management empowered by reinforcement learning,

    T. Zhang, A. Hellander, and S. Toor, “Efficient hierarchical stor- age management empowered by reinforcement learning,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 6, pp. 5780–5793, 2022

    2022

  57. [66]

    Reinforcement learning-enhanced shared-account cross-domain sequential recom- mendation,

    L. Guo, J. Zhang, T. Chen, X. Wang, and H. Yin, “Reinforcement learning-enhanced shared-account cross-domain sequential recom- mendation,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 7, pp. 7397–7411, 2022

    2022

  58. [67]

    Learning from atypical behavior: temporary interest aware recommendation based on reinforcement learning,

    Z. Du, N. Yang, Z. Yu, and S. Y. Philip, “Learning from atypical behavior: temporary interest aware recommendation based on reinforcement learning,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 10, pp. 9824–9835, 2022

    2022

  59. [68]

    Large language models are semi-parametric reinforcement learning agents,

    D. Zhang, L. Chen, S. Zhang, H. Xu, Z. Zhao, and K. Yu, “Large language models are semi-parametric reinforcement learning agents,”Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  60. [69]

    Multi-agent reinforcement learning is a sequence modeling prob- lem,

    M. Wen, J. Kuba, R. Lin, W. Zhang, Y. Wen, J. Wang, and Y. Yang, “Multi-agent reinforcement learning is a sequence modeling prob- lem,”Advances in Neural Information Processing Systems, vol. 35, pp. 16 509–16 521, 2022

    2022

  61. [70]

    Automated feature selection: A reinforcement learning perspective,

    K. Liu, Y. Fu, L. Wu, X. Li, C. Aggarwal, and H. Xiong, “Automated feature selection: A reinforcement learning perspective,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 3, pp. 2272–2284, 2021

    2021

  62. [71]

    Mo-mix: Multi-objective multi-agent cooperative decision-making with deep reinforcement learning,

    T. Hu, B. Luo, C. Yang, and T. Huang, “Mo-mix: Multi-objective multi-agent cooperative decision-making with deep reinforcement learning,”IEEE Transactions on Pattern Analysis and Machine Intelli- gence, vol. 45, no. 10, pp. 12 098–12 112, 2023

    2023

  63. [72]

    Traceable group-wise self-optimizing feature transformation learning: A dual optimization perspective,

    M. Xiao, D. Wang, M. Wu, K. Liu, H. Xiong, Y. Zhou, and Y. Fu, “Traceable group-wise self-optimizing feature transformation learning: A dual optimization perspective,”ACM Transactions on Knowledge Discovery from Data, vol. 18, no. 4, pp. 1–22, 2024

    2024

  64. [73]

    Comprehend, divide, and conquer: Feature subspace exploration via multi-agent hierarchical reinforcement learning,

    W. Zhang, X. Huang, Y. Du, Z. Qiao, Q. Long, Z. Meng, Y. Zhou, and M. Xiao, “Comprehend, divide, and conquer: Feature subspace exploration via multi-agent hierarchical reinforcement learning,” arXiv preprint arXiv:2504.17356, 2025. Meng Xiaoreceived joint doctoral training fro...

    2025 arXiv

  65. [77]

    supervisor, the deputy director of Computer Network Information Center, Chinese Academy of Sciences

    He is a professor, Ph.D. supervisor, the deputy director of Computer Network Information Center, Chinese Academy of Sciences. His re- search interests include data mining, big data pro- cessing, and knowledge graph. He has published prolifically in refereed journals and confer...

  66. [2006]

    He also won the CVPR UG2+ challenge in 2021 and the IJCAI competi- tion on repeated buyers prediction in 2015

    He received the best paper awards in EMBS Society 2023, IEEE ICIEA 2022, IEEE SmartCity 2022, InCoB 2016 and DASFAA 2015. He also won the CVPR UG2+ challenge in 2021 and the IJCAI competi- tion on repeated buyers prediction in 2015. He has been serving as an Associate Editor f...

    2023

  67. [2020]

    Available: http://dx.doi.org/10.1093/bib/bbaa306

    [Online]. Available: http://dx.doi.org/10.1093/bib/bbaa306

    doi

  68. [2021]

    Available: http://dx.doi.org/10.1093/bib/bbab034

    [Online]. Available: http://dx.doi.org/10.1093/bib/bbab034

    doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.