Pith. sign in

REVIEW 4 major objections 5 minor 71 references

Neurosymbolic Learning for Predicting Cell Fate Decisions from Longitudinal Single Cell Transcriptomics in Paediatric Acute Myeloid Leukemia

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read This paper claims that combining deep temporal models with algorithmic complexity measurements on longitudinal single-cell transcriptomes identifies plasticity markers that predict paediatric AML cell-fate transitions and expose stalled-dif

desk verdict Claims predictive biomarkers for pediatric AML but never evaluates whether any model predicts—feature importance is not forecast. read the letter →

arxiv 2508.13199 v1 pith:NOVOA5FX submitted 2025-08-16 q-bio.QM

classification q-bio.QM
keywords paediatricacutemyeloidleukemiasingle-cellRNAsequencingcellfateplasticityalgorithmicinformationdynamicsrecurrentneuralnetworkstrajectoryinferenceattractorlandscapedifferentiationtherapy
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 sets out to show that the trajectory of paediatric acute myeloid leukaemia—from diagnosis to remission to relapse—can be predicted from single-cell gene expression by combining recurrent and attention-based neural networks with algorithmic information dynamics, a complexity measure that scores how much each gene or interaction contributes to the structure of the gene regulatory network. Applied to pseudobulk longitudinal scRNA-seq from 14 patients, the framework identifies a small set of "plasticity markers" whose perturbation scores and neural-network importance ranks are consistently high, and interprets them as bifurcation signatures that steer cell fate decisions. The authors use these signatures to argue that AML cells are not fixed leukemic stem-cell clones but occupy stalled, reprogrammable attractor states: developmental arrest blocks terminal differentiation, and the inferred state space mixes haematopoietic, lymphoid-like, erythroid, and neurodevelopmental programs. If true, the same markers become predictive biomarkers for relapse and concrete targets for differentiation therapy—WNT modulation, histone demethylase inhibition, or CRISPR-based reprogramming.

What carries the argument

The load-bearing object is the Block Decomposition Method (BDM) perturbation analysis, a graph-complexity estimator that approximates Kolmogorov complexity by decomposing gene correlation networks into small blocks and summing their algorithmic probabilities. Node- and link-based perturbations measure how much network complexity drops when a gene or interaction is removed, yielding a causal-influence score beyond what correlation, entropy, or centrality capture. This BDM signal is combined with feature-importance rankings from LSTM, BiLSTM, and Transformer models trained on the three clinical timepoints; the union of these rankings defines the plasticity signatures that the paper treats as r

What would settle it

Collect a dense time series (or lineage tracing) across relapse in a validation cohort and test whether the top BDM-shifted genes change expression state before relapse; then perturb them with CRISPR knockdown or activation and measure whether differentiation trajectories in AML cells shift accordingly. Negative results on either leg would refute the causal-plasticity claim.

Watch

Extended reading notes

Core claim

The central claim is that causal drivers of AML plasticity can be discovered by integrating deep learning with algorithmic information dynamics, without needing multi-omic input. Using longitudinal single-cell transcriptomes at diagnosis, remission, and relapse, the authors identify a convergent set of transition genes—H3F3A/B, KDM5B, TPT1, TLE1, S100A9, KCNE5, CD14, and others—as plasticity markers regulating cell fate bifurcations. They interpret these signatures as evidence that AML cells are arrested near a common myeloid progenitor-like attractor with biased megakaryocyte–erythroid and granulocyte–monocyte lineage branches, and that disrupted bioelectric, immune, and epigenetic signalin

Load-bearing premise

The load-bearing premise is that three clinical snapshots (diagnosis, remission, relapse) from 14 patients, after pseudobulk aggregation, are enough to reconstruct AML's dynamic state space; if sparse sampling or batch effects dominate, the inferred transitions and attractors are artifacts.

Editorial extensions

If this is right

  • The 20–30 consistently top-ranked genes are proposed as causal regulators of AML plasticity, not merely statistical correlates.
  • Targeting these markers—via WNT activation with GSK3β inhibition, KDM5B or LSD1 inhibition, H3F3A/B knockdown, TPT1 deletion, or KLF1 activation—should push leukemic cells out of arrested states toward terminal differentiation.
  • The same signatures can serve as relapse-risk biomarkers, since they shift between diagnosis, remission, and relapse and could be monitored in liquid biopsies for early recurrence detection.
  • The overlap between AML plasticity markers and paediatric high-grade glioma signatures implies a shared developmental program, making cross-cancer differentiation-therapy strategies plausible.
  • AML cell-fate decisions appear biased near a common myeloid progenitor branch point with megakaryocyte–erythroid and granulocyte–monocyte lineages, rather than being purely stochastic.

Reading between the lines

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

  • If the three-state representation is faithful, the diagnosis-to-remission changes alone should already forecast relapse; this is a holdout prediction the paper does not report, and a direct way to test the framework's causal claim.
  • The neurodevelopmental/ectoderm–mesoderm reading suggests a concrete experiment: ask whether neural-crest transcription factors (SOX10, DLX6, POU3F2) drive the hybrid identity in AML cell lines, e.g., via CRISPRa of these factors followed by scRNA-seq trajectory inference.
  • The BDM perturbation scores could be benchmarked against simple network centralities and against random gene deletions; if BDM adds no predictive power, the specific algorithmic-complexity contribution would be weakened.
  • Pseudobulk aggregation erases cell-level heterogeneity, so a natural extension is to apply the same pipeline to single-cell pseudo-time ordering to see whether the same plasticity markers appear within individual clones.
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

4 major / 5 minor

Summary. The paper proposes a 'neurosymbolic' framework that combines Algorithmic Information Dynamics (BDM perturbation), NNMF clustering, and deep learning models (LSTM, BiLSTM, Transformer, Bayesian Ridge Regression) applied to public bulk and longitudinal single-cell AML transcriptomic data. From a random subset of 14 of 28 patients with three clinical timepoints (diagnosis, remission, relapse), the authors derive DEGs, BDM perturbation scores, and feature-importance rankings, and interpret convergence across methods as the discovery of predictive plasticity biomarkers and causal regulators of AML cell fate transitions. The abstract and conclusions further claim that these signatures reveal ectoderm–mesoderm crosstalk, a brain–immune–hematopoietic axis, and actionable differentiation-therapy targets. The central scientific claim is that the models 'predict' or 'forecast' AML cell fate transitions and identify causal biomarkers.

Significance. If the central claim were established, a validated longitudinal deep-learning framework for predicting AML relapse from single-cell transcriptomics would be a meaningful contribution to computational oncology because true predictive biomarkers for paediatric AML relapse are an unmet clinical need. The manuscript has strengths: it uses publicly available datasets, provides a GitHub code repository, and attempts to combine several complementary methods rather than a single black-box model. However, the evidence presented does not support the central claim: no predictive performance is reported, no held-out validation is described, and the causal language greatly exceeds what the correlation- and perturbation-based analyses can establish. The contribution is therefore best viewed as an exploratory signature-discovery pipeline whose translational and predictive claims remain unsubstantiated.

major comments (4)
  1. [Methods (LSTM, BiLSTM, Transformer, BRR) and Results (AI-Driven Longitudinal Biomarker Discovery; Figure 3)] The paper never reports any measure of predictive performance for the deep learning models. The Methods describe a 70/15/15 train/validation/test split for classification across DX/REL/REM states, but the Results contain no accuracy, sensitivity, specificity, AUC, or any held-out test prediction. Feature-importance weights extracted from trained model parameters are not equivalent to predictive accuracy; a model can assign high importance to features while still failing to generalize. Because the abstract's central claim is that the work 'identifies key plasticity markers as predictive signatures' and that the algorithms 'forecast' cell fate transitions, the absence of any predictive validation is load-bearing and unsupported.
  2. [Datasets and Preprocessing; BDM Analysis; AI-Driven Longitudinal Biomarker Discovery] There is a circularity in the biomarker discovery. The top-100 DEGs are selected from the same 14-patient pseudobulk cohort that is then used to train the deep learning models and to construct the BDM correlation networks. The feature rankings extracted from those models are subsequently presented as convergent evidence for the same genes. No held-out patients or independent cohort are used. The 14 patients are chosen 'randomly' from 28 without a reported seed, and the manuscript does not demonstrate that this subset is representative. Consequently, overlap between LSTM, Transformer, BDM, and NNMF gene lists may reflect shared input features and thresholds rather than biological convergence.
  3. [BDM Analysis; Results (Single-Cell BDM Perturbation Signatures); Discussion] The causal claims are not supported by the methodology. BDM perturbation is applied to adjacency matrices constructed from Spearman correlations and binarized at arbitrary thresholds (0.5 or 0.1). Deleting a node and measuring the change in algorithmic complexity of a correlation graph does not establish that the gene is a causal regulator of cell fate, nor does it identify a 'critical tipping point' or 'causal control point.' Bayesian Ridge regression coefficients are also associative. The manuscript repeatedly uses 'causal drivers,' 'causal biomarkers,' and 'critical transition genes' in the Discussion and Conclusions, and Table 3 proposes CRISPR and small-molecule therapeutic combinations based on these rankings. Without perturbation experiments, causal or therapeutic claims exceed the evidence.
  4. [Datasets and Preprocessing; Limitations and Mechanistic Interpretations] The trajectory-inference and forecasting claims are undermined by the data structure. Expression is pseudobulk-aggregated from three clinical snapshots per patient, so the models observe three coarse timepoints rather than continuous cell fate transitions. The Limitations section correctly concedes that 'sparse clinical timepoints' constrain trajectory resolution and that multi-omic validation is a future priority, but the Conclusions nevertheless state that the algorithms 'forecast' cell fate trajectories and identify causal biomarkers. The manuscript does not control for batch or technical effects beyond log-normalization, and the 14-patient subset is not shown to preserve subtype diversity or temporal completeness in any quantitative way.
minor comments (5)
  1. [Results (first paragraph)] Typo: 'Dabatase' should be 'Database.' Also, Figure 2 panels E-G are described as scatter-like plots but the axes are not fully defined; please clarify what 'count frequency' and 'average BDM perturbation' represent.
  2. [Methods (LSTM, Transformer)] No random seed, hyperparameter tuning, early stopping, or regularization strategy beyond dropout is reported for the deep learning models. With only 14 patients and 50 epochs, overfitting is a serious concern; please report seeds, tuning procedure, and any regularization used.
  3. [Conclusions; Table 3] Table 3 lists numerous therapeutic combinations (e.g., KDM5B inhibition, WNT activation, TPT1 deletion) without any experimental validation. These should be clearly labeled as speculative hypotheses, not 'precision targets' as implied in the text.
  4. [References] References [61] and [62] appear to be the same STGRNS entry and should be consolidated.
  5. [Introduction; Conclusions] Typo: 'linage' should be 'lineage'; in the Conclusions, 'PCHD1/2' should likely be 'PCDHA1/2.'

Circularity Check

2 steps flagged · score 6.0 of 10

In-sample feature importance and BDM re-ranking of the same DEGs are presented as validated predictive biomarkers; no held-out prediction is reported.

  1. fitted input called prediction [Methods (LSTM Analysis; Transformer Analysis) and Results ('AI-Driven Longitudinal Biomarker Discovery', Figure 3)]
    "Feature importance was computed from the LSTM input weights matrix by taking the mean absolute value across gates/units. The top 100 genes ranked by feature importance were extracted and subjected to BDM perturbation analysis. ... All three longitudinal AI algorithms forecasted epigenetic signals predicting AML cell fate trajectories."

    The LSTM/Transformer models are trained on the same 14 patients' DX/REM/REL pseudobulk trajectories from which the 'predictive' feature importances are extracted. The paper reports no held-out test accuracy, AUC, sensitivity, or external cohort validation anywhere. The 'forecasts' are therefore the in-sample input-weight and attention averages of models fitted to the very data being 'predicted.' Calling these fitted statistics 'predictive signatures' is a renaming of training outputs as prediction, not an independent predictive result.

  2. self definitional [Methods (DEG Extraction and Statistical Analysis; BDM Analysis) and Results ('Single-Cell BDM Perturbation Signatures Reveal Plasticity Regulators...', Figure 2E)]
    "From these, the top 100 DEGs, ranked by absolute LogFC, were selected for downstream analysis to capture longitudinal state-transition dynamics, and predict plasticity regulators steering the cell fate trajectories. ... adjacency matrices were first constructed from the Spearman correlation coefficients between DEGs. ... Figure 2, panel E highlights the top DEGs with the highest positive perturbed BDM shifts across the AML scRNA-Seq data."

    The genes later reported as 'plasticity regulators' and 'bifurcation signatures' are, by construction, members of the same top-100 DEG list used to build the BDM correlation networks. The BDM perturbation score only re-ranks these input genes; no independent gene set, functional perturbation, or held-out evaluation is introduced. Thus the 'identification' of plasticity markers reduces to a re-labelling of the initial DEG-selection step rather than an independent discovery.

full rationale

The paper's central claim—that it identifies predictive plasticity markers regulating AML cell fate transitions—is built in a closed analysis loop. First, the deep learning models are fit to the same 14-patient longitudinal dataset from which feature importance is read off the fitted weights; with no held-out or external prediction reported, the 'predictive signatures' are in-sample summary statistics presented as forecasts. Second, the BDM 'causal control points' are perturbations of correlation networks whose nodes are exactly the DEGs selected in the preceding step, so the top-BDM genes are guaranteed to be members of the DEG input list; calling them independently discovered plasticity regulators is a renaming of the input selection criterion. The self-citations to AID/BDM methods [55,56,54,67,66,68] are not additionally counted as circular here because those are published, code-reproduced methods whose assumptions do not include the AML result; they are real prior evidence, though they do not supply the missing validation. The score is 6 rather than higher because the computations themselves are not literally identical to the input—a properly held-out design or independent functional assay could in principle give the feature rankings independent content. As written, however, the 'predictive biomarker' discovery is forced by the construction of the pipeline.

Assumptions & free parameters 4 free parameters · 5 assumptions · 3 invented entities

The analysis rests on several untested modeling choices. The BDM perturbation method assumes correlation networks binarized at arbitrary thresholds reflect causal regulatory structure. The deep learning feature importance is taken as biological significance without validation. The sparse three-timepoint sampling from 14 patients is assumed sufficient to infer cell fate trajectories.

free parameters (4)
  • BDM binarization threshold = 0.5 (and 0.1 in some panels)
    Used to convert Spearman correlation matrices to graphs; chosen by hand without sensitivity analysis, directly affects BDM perturbation scores.
  • NNMF component count = 5
    Number of co-expression modules selected without justification; affects module membership and downstream BDM shifts.
  • Top-k gene cutoffs = 100 DEGs, 50 genes/module, 100 feature-importance genes
    Arbitrary cutoffs that determine which genes enter network analyses; no stability analysis.
  • Patient subset (14 of 28) = 14 patients
    Randomly selected half the cohort for computational reasons; no seed or representativeness check, could bias results.
assumptions (5)
  • domain assumption Binarized Spearman correlation networks at threshold 0.5 represent gene regulatory networks.
    Methods BDM Analysis; no evidence that correlation structure aligns with regulatory edges.
  • domain assumption BDM perturbation scores measure causal influence of nodes and links.
    Methods BDM Analysis; the causal interpretation is asserted from prior AID work without benchmark validation.
  • domain assumption Feature importance from LSTM/Transformer weights reflects biological importance.
    Methods LSTM/BiLSTM/Transformer; no testing against known regulators or perturbations.
  • domain assumption Three clinical timepoints (DX, REM, REL) suffice to infer cell fate trajectories and attractors.
    Methods Datasets and Preprocessing; longitudinal snapshots are treated as a continuous trajectory without lineage tracing or dense sampling.
  • domain assumption Pseudobulk aggregation preserves trajectory information.
    Methods Datasets and Preprocessing; averaging over cells within each timepoint discards single-cell heterogeneity central to the stated goal.
invented entities (3)
  • Ectoderm-mesoderm crosstalk
    purpose: Proposed mechanism linking neurodevelopmental and hematopoietic gene signatures to AML plasticity
    Inferred solely from co-enrichment of neural and hematopoietic genes in BDM/BRR lists; no functional assay or lineage tracing supports it.
  • Brain-immune-hematopoietic axis
    purpose: Interpretation of neurotransmitter and immune gene signals as a systemic communication axis in AML
    Based on enrichment of receptor/transporter genes; no direct evidence of brain-immune interaction in the patient data.
  • Bioelectric signaling network
    purpose: Explanation of ion channel and solute carrier gene perturbations as regulators of cell fate
    Ion channel genes show BDM shifts, but the claim that they 'steer' fate decisions is not tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neurosymbolic Learning for Predicting Cell Fate Decisions from Longitudinal Single Cell Transcriptomics in Paediatric Acute Myeloid Leukemia." pith.science (2026). https://pith.science/paper/NOVOA5FX

@misc{pith2026250813199,
  author       = {Pith},
  title        = {Pith review of: Neurosymbolic Learning for Predicting Cell Fate Decisions from Longitudinal Single Cell Transcriptomics in Paediatric Acute Myeloid Leukemia},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NOVOA5FX}},
  note         = {Machine review of arXiv:2508.13199}
}
read the original abstract

Paediatric Acute Myeloid Leukemia is a complex adaptive ecosystem with high morbidity. Current trajectory inference algorithms struggle to predict causal dynamics in AML progression, including relapse and recurrence risk. We propose a symbolic AI and deep learning framework grounded in complexity science, integrating Recurrent Neural Networks, Transformers, and Algorithmic Information Dynamics to model longitudinal single cell transcriptomics and infer complex state transitions in paediatric AML. We identify key plasticity markers as predictive signatures regulating developmental trajectories. These were derived by integrating deep learning with complex systems based network perturbation analysis and dynamical systems theory to infer high dimensional state space attractors steering AML evolution. Findings reveal dysregulated epigenetic and developmental patterning, with AML cells in maladaptive, reprogrammable plastic states, i.e., developmental arrest blocking terminal differentiation. Predictions forecast neurodevelopmental and morphogenetic signatures guiding AML cell fate bifurcations, suggesting ectoderm mesoderm crosstalk during disrupted differentiation. Neuroplasticity and neurotransmission related transcriptional signals implicate a brain immune hematopoietic axis in AML cell fate cybernetics. This is the first study combining RNNs and AID to predict and decipher longitudinal patterns of cell fate transition trajectories in AML. Our complex systems approach enables causal discovery of predictive biomarkers and therapeutic targets for ecosystem engineering, cancer reversion, precision gene editing, and differentiation therapy, with strong translational potential for precision oncology and patient centered care.

Figures

Figures reproduced from arXiv: 2508.13199 by the authors.

Figure 1
Figure 1. Symbolic deep learning algorithms to predict AML state￾transitions and deconstruct longitudinal single-cell transcriptional landscapes. The schematic outlines a four-step workflow starting from single￾cell gene expression matrices across diagnosis (DX), relapse (REL), and remis￾sion (REM) AML states. It integrates differential gene and module detection, followed by deep learning-based temporal classification, and BD… view at source ↗
Figure 2
Figure 2. Top 20 Spearman and Pearson Correlation Heatmaps for Healthy and [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Panels A–G illustrate the top 30 genes with the highest importance [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 59 canonical work pages

  1. [1]

    Transition Therapy: Tackling the Ecology of Tumor Phenotypic Plasticity

    G. Aguad´ e-Gorgor ´ ıo, S. Kauffman, and R. Sol´ e. “Transition Therapy: Tackling the Ecology of Tumor Phenotypic Plasticity”. In: Bulletin of Mathematical Biology 84.24 (2022). doi: 10.1007/s11538-021-00970-9

  2. [2]

    The role of HOX genes in normal hematopoiesis and acute leukemia

    Rasha A Alharbi et al. “The role of HOX genes in normal hematopoiesis and acute leukemia”. In: Leukemia 27.5 (2013), pp. 1000–1008. doi: 10. 1038/leu.2012.356. url: https://doi.org/10.1038/leu.2012.356

  3. [3]

    Differentiating Neuroblastoma: A Systematic Review of the Retinoic Acid, Its Derivatives, and Synergistic Interactions

    N. Bayeva, E. Coll, and O. Piskareva. “Differentiating Neuroblastoma: A Systematic Review of the Retinoic Acid, Its Derivatives, and Synergistic Interactions”. In: Journal of Personalized Medicine 11.3 (2021), p. 211. doi: 10.3390/jpm11030211

  4. [4]

    Redifferentiation therapeutic strategies in cancer

    M. Bizzarri et al. “Redifferentiation therapeutic strategies in cancer”. In: Drug Discovery Today 25.4 (Apr. 2020), pp. 731–738. doi: 10.1016/j. drudis.2020.01.021

  5. [5]

    Mutant H3 histones drive human pre-leukemic hematopoi- etic stem cell expansion and promote leukemic aggressiveness

    M. Boileau et al. “Mutant H3 histones drive human pre-leukemic hematopoi- etic stem cell expansion and promote leukemic aggressiveness”. In: Nature Communications 10.1 (2019), p. 2891. doi: 10.1038/s41467-019-10705- z

  6. [6]

    The molecular landscape of pediatric acute myeloid leukemia reveals recurrent structural alterations and age-specific muta- tional interactions

    H. Bolouri et al. “The molecular landscape of pediatric acute myeloid leukemia reveals recurrent structural alterations and age-specific muta- tional interactions”. In: Nature Medicine 24.1 (2018), pp. 103–112. doi: 10.1038/nm.4439. url: https://doi.org/10.1038/nm.4439. 32

  7. [7]

    New data on robustness of gene expression sig- natures in leukemia: comparison of three distinct total RNA preparation procedures

    M. Campo Dell’Orto et al. “New data on robustness of gene expression sig- natures in leukemia: comparison of three distinct total RNA preparation procedures”. In: BMC Genomics 8 (2007), p. 188. doi: 10.1186/1471- 2164-8-188

  8. [8]

    Multi-omics reveals mito- chondrial metabolism proteins susceptible for drug discovery in AML

    M. Caplan, K. J. Wittorf, K. K. Weber, et al. “Multi-omics reveals mito- chondrial metabolism proteins susceptible for drug discovery in AML”. In: Leukemia 36 (2022), pp. 1296–1305. doi: 10.1038/s41375-022-01518-z . url: https://doi.org/10.1038/s41375-022-01518-z

Show all 71 references
  1. [9]

    Transformer for one-stop interpretable cell type annotation

    J. Chen, H. Xu, and W. Tao. “Transformer for one-stop interpretable cell type annotation”. In: Nature Communications 14 (2023), p. 223. doi: 10.1038/s41467-023-35923-4

  2. [10]

    Inferring gene regulatory networks from time-series scRNA-seq data via GRANGER causal recurrent autoencoders

    Liang Chen et al. “Inferring gene regulatory networks from time-series scRNA-seq data via GRANGER causal recurrent autoencoders”. In:Brief- ings in Bioinformatics 26.2 (2025). doi: 10 . 1093 / bib / bbaf089. url: https://doi.org/10.1093/bib/bbaf089

  3. [11]

    Developmental origins shape the paediatric cancer genome

    Xiaolong Chen et al. “Developmental origins shape the paediatric cancer genome”. In: Nature Reviews Cancer 24.6 (2024), pp. 382–398. doi: 10. 1038/s41568-024-00684-9 . url: https://doi.org/10.1038/s41568- 024-00684-9

  4. [12]

    Chromatin accessibility: biological func- tions, molecular mechanisms and therapeutic application

    Y. Chen, R. Liang, Y. Li, et al. “Chromatin accessibility: biological func- tions, molecular mechanisms and therapeutic application”. In:Signal Trans- duction and Targeted Therapy 9 (2024), p. 340. doi: 10.1038/s41392- 024-02030-9. url: https://doi.org/10.1038/s41392-024-02030-9

  5. [13]

    Identifying critical transitions in complex diseases

    S. Deb, S. Bhandary, S. K. Sinha, et al. “Identifying critical transitions in complex diseases”. In: Journal of Biosciences 47 (2022), p. 25. doi: 10 . 1007 / s12038 - 022 - 00258 - 7. url: https : / / doi . org / 10 . 1007 / s12038-022-00258-7

  6. [14]

    Lineage switching in acute leukemias: a consequence of stem cell plasticity?

    E. Dorantes-Acosta and R. Pelayo. “Lineage switching in acute leukemias: a consequence of stem cell plasticity?” In: Bone Marrow Research 2012 (2012), p. 406796. doi: 10.1155/2012/406796

  7. [15]

    535 Multiomic analysis of KMT2A-r pediatric acute myeloid leukemia

    T. Edwards et al. “535 Multiomic analysis of KMT2A-r pediatric acute myeloid leukemia”. In:Journal of Clinical and Translational Science9.Suppl 1 (2025), p. 157. doi: 10.1017/cts.2024.1111. url: https://doi.org/ 10.1017/cts.2024.1111

  8. [16]

    Aberrant DNA methylation at HOXA4 and HOXA5 genes are associated with resistance to imatinib mesylate among chronic myeloid leukemia patients

    M.H. Elias et al. “Aberrant DNA methylation at HOXA4 and HOXA5 genes are associated with resistance to imatinib mesylate among chronic myeloid leukemia patients”. In: Cancer Reports 1.2 (2018), e1111. doi: 10.1002/cnr2.1111. url: https://doi.org/10.1002/cnr2.1111

  9. [17]

    State-transition mod- eling of blood transcriptome predicts disease evolution and treatment re- sponse in chronic myeloid leukemia

    D.E. Frankhouser, R.C. Rockne, L. Uechi, et al. “State-transition mod- eling of blood transcriptome predicts disease evolution and treatment re- sponse in chronic myeloid leukemia”. In:Leukemia 38 (2024), pp. 769–780. doi: 10.1038/s41375-024-02142-9 . 33

  10. [18]

    Reinforcement learning guides single-cell sequencing in decoding lineage and cell fate decisions

    Zeyu Fu et al. “Reinforcement learning guides single-cell sequencing in decoding lineage and cell fate decisions”. In: bioRxiv (2024). doi: 10 . 1101/2024.07.04.602019

  11. [19]

    Control of Cellular Differentiation Trajectories for Can- cer Reversion

    J. R. Gong et al. “Control of Cellular Differentiation Trajectories for Can- cer Reversion”. In: Advanced Science 12.3 (Jan. 2025). Epub 2024 Dec 11, e2402132. doi: 10.1002/advs.202402132

  12. [20]

    Quantifying cancer cell plasticity with gene regulatory networks and single-cell dynamics

    S. M. Groves and V. Quaranta. “Quantifying cancer cell plasticity with gene regulatory networks and single-cell dynamics”. In: Frontiers in Net- work Physiology 3 (2023), p. 1225736. doi: 10 . 3389 / fnetp . 2023 . 1225736. url: https://doi.org/10.3389/fnetp.2023.1225736

  13. [21]

    Epigenetic reprogramming in pediatric gliomas: from molecular mechanisms to therapeutic implications

    S. Haase et al. “Epigenetic reprogramming in pediatric gliomas: from molecular mechanisms to therapeutic implications”. In: Trends in Can- cer 10.12 (2024), pp. 1147–1160. doi: 10.1016/j.trecan.2024.09.007 . url: https://doi.org/10.1016/j.trecan.2024.09.007

  14. [22]

    Hallmarks of Cancer: New Dimensions

    D. Hanahan. “Hallmarks of Cancer: New Dimensions”. In: Cancer Dis- covery 12.1 (2022), pp. 31–46. doi: 10.1158/2159- 8290.CD- 21- 1059. url: https://doi.org/10.1158/2159-8290.CD-21-1059

  15. [23]

    Cancer attractors: a systems view of tumors from a gene network dynamics and developmental perspective

    S. Huang, I. Ernberg, and S. Kauffman. “Cancer attractors: a systems view of tumors from a gene network dynamics and developmental perspective”. In: Seminars in Cell & Developmental Biology 20.7 (2009), pp. 869–876. doi: 10.1016/j.semcdb.2009.07.003 . url: https://doi.org/10. ...

  16. [24]

    Retinoic acid induces differentiation in neuroblastoma via ROR1 by modulating retinoic acid response elements

    A. Illendula, N. Fultang, and B. Peethambaran. “Retinoic acid induces differentiation in neuroblastoma via ROR1 by modulating retinoic acid response elements”. In: Oncology Reports 44.3 (2020), pp. 1013–1024. doi: 10.3892/or.2020.7681

  17. [25]

    Stalled developmental programs at the root of pediatric brain tumors

    S. Jessa et al. “Stalled developmental programs at the root of pediatric brain tumors”. In: Nature Genetics 51 (2019), pp. 1702–1713. doi: 10. 1038/s41588-019-0531-7 . url: https://doi.org/10.1038/s41588- 019-0531-7

  18. [26]

    ScLSTM: single-cell type detection by siamese recurrent network and hierarchical clustering

    H. Jiang et al. “ScLSTM: single-cell type detection by siamese recurrent network and hierarchical clustering”. In:BMC Bioinformatics 24.1 (2023), p. 417. doi: 10.1186/s12859-023-05494-8

  19. [27]

    Characterization of cell-fate decision landscapes by estimating transcription factor dynamics

    S. Jim´ enez et al. “Characterization of cell-fate decision landscapes by estimating transcription factor dynamics”. In: Cell Reports Methods 3.7 (2023), p. 100512. doi: 10.1016/j.crmeth.2023.100512

  20. [28]

    Data-driven modeling of core gene regulatory network underlying leukemogenesis in IDH mutant AML

    A. Katebi et al. “Data-driven modeling of core gene regulatory network underlying leukemogenesis in IDH mutant AML”. In: npj Systems Biology and Applications 10 (2024), p. 38. doi: 10.1038/s41540-024-00366-0

  21. [29]

    A longitudinal single-cell atlas of treatment response in pediatric AML

    S. Lambo et al. “A longitudinal single-cell atlas of treatment response in pediatric AML”. In: Cancer Cell 41.12 (2023), 2117–2135.e12. doi: 10.1016/j.ccell.2023.10.008. 34

  22. [30]

    Transformer-Based Single-Cell Language Model: A Sur- vey

    W. Lan et al. “Transformer-Based Single-Cell Language Model: A Sur- vey”. In: Big Data Mining and Analytics 7.4 (2024), pp. 1169–1186. doi: 10.26599/BDMA.2024.9020034

  23. [31]

    Reconstructing the regulatory programs underlying the phenotypic plasticity of neural cancers

    I. Larsson et al. “Reconstructing the regulatory programs underlying the phenotypic plasticity of neural cancers”. In: Nature Communications 15.1 (2024), p. 9699. doi: 10 . 1038 / s41467 - 024 - 53954 - 3. url: https : //doi.org/10.1038/s41467-024-53954-3

  24. [32]

    Decoding the principle of cell-fate de- termination for its reverse control

    J. Lee, N. Kim, and K. H. Cho. “Decoding the principle of cell-fate de- termination for its reverse control”. In: NPJ Systems Biology and Appli- cations 10.1 (2024), p. 47. doi: 10 . 1038 / s41540 - 024 - 00372 - 2. url: https://doi.org/10.1038/s41540-024-00372-2

  25. [33]

    Bioelectric networks: the cognitive glue enabling evo- lutionary scaling from physiology to mind

    Michael Levin. “Bioelectric networks: the cognitive glue enabling evo- lutionary scaling from physiology to mind”. In: Animal Cognition 26.6 (2023), pp. 1865–1891. doi: 10.1007/s10071-023-01780-3

  26. [34]

    Dynamics inside the cancer cell attractor reveal cell het- erogeneity, limits of stability, and escape

    Q. Li et al. “Dynamics inside the cancer cell attractor reveal cell het- erogeneity, limits of stability, and escape”. In: Proceedings of the Na- tional Academy of Sciences of the United States of America 113.10 (2016), pp. 2672–2677. doi: 10.1073/pnas.1519210113. url: https:/...

  27. [35]

    Attention-based deep clustering method for scRNA-seq cell type identification

    S. Li et al. “Attention-based deep clustering method for scRNA-seq cell type identification”. In: PLoS Comput Biol 19.11 (2023), e1011641. doi: 10.1371/journal.pcbi.1011641

  28. [36]

    Aberrant stem cell and developmental programs in pediatric leukemia

    Rachael E Ling, Jennifer W Cross, and Ananya Roy. “Aberrant stem cell and developmental programs in pediatric leukemia”. In: Frontiers in Cell and Developmental Biology 12 (2024), p. 1372899. doi: 10.3389/fcell. 2024.1372899

  29. [37]

    scLEGA: an attention-based deep clustering method with a tendency for low expression of genes on single-cell RNA-seq data

    Zhenze Liu et al. “scLEGA: an attention-based deep clustering method with a tendency for low expression of genes on single-cell RNA-seq data”. In: Briefings in Bioinformatics 25 (5 2024). doi: 10.1093/bib/bbae371

  30. [38]

    Cell Fate Decision as High-Dimensional Critical State Transition

    M. Mojtahedi et al. “Cell Fate Decision as High-Dimensional Critical State Transition”. In: PLoS Biology 14.12 (2016), e2000640. doi: 10 . 1371 / journal.pbio.2000640 . url: https://doi.org/10.1371/journal. pbio.2000640

  31. [39]

    Single-cell analysis reveals altered tumor microen- vironments of relapse- and remission-associated pediatric acute myeloid leukemia

    H. Mumme et al. “Single-cell analysis reveals altered tumor microen- vironments of relapse- and remission-associated pediatric acute myeloid leukemia”. In: Nature Communications 14.1 (2023), p. 6209. doi: 10 . 1038/s41467-023-41994-0

  32. [40]

    Identification of leukemia- enriched signature through the development of a comprehensive pediatric single-cell atlas

    H.L. Mumme, C. Huang, D. Ohlstrom, et al. “Identification of leukemia- enriched signature through the development of a comprehensive pediatric single-cell atlas”. In: Nature Communications 16.1 (2025), p. 4114. doi: 10.1038/s41467-025-59362-5 . 35

  33. [41]

    PDQ Childhood Cancer Ge- nomics

    PDQ Pediatric Treatment Editorial Board. PDQ Childhood Cancer Ge- nomics. https://www.cancer.gov/types/childhood-cancers/pediatric- genomics- hp- pdq. Updated 07/28/2025. Accessed 07/28/2025. [PMID: 27466641]. 2025

  34. [42]

    Causality: Models, Reasoning and Inference

    Judea Pearl. Causality: Models, Reasoning and Inference. 2nd. New York, NY, United States: Cambridge University Press, 2009, p. 478. isbn: 978- 0-521-89560-6

  35. [43]

    Tumor reversion and embryo morphogenetic factors

    S. Proietti, A. Cucina, and A. Pensotti. “Tumor reversion and embryo morphogenetic factors”. In: Seminars in Cancer Biology 79 (Feb. 2022). Epub 2020 Sep 10, pp. 83–90. doi: 10.1016/j.semcancer.2020.09.005

  36. [44]

    FateNet: an integration of dy- namical systems and deep learning for cell fate prediction

    Mehrshad Sadria and Thomas M. Bury. “FateNet: an integration of dy- namical systems and deep learning for cell fate prediction”. In: Bioinfor- matics 40.9 (2024), btae525. doi: 10.1093/bioinformatics/btae525

  37. [45]

    H3K27M mutant glioma: Disease definition and bio- logical underpinnings

    AM Saratsis et al. “H3K27M mutant glioma: Disease definition and bio- logical underpinnings”. In: Neuro-Oncology 26.Supplement 2 (2024), S92– S100. doi: 10.1093/neuonc/noad164

  38. [46]

    Anticipating critical transitions in epithelial-hybrid-mesenchymal cell-fate determination

    S. Sarkar et al. “Anticipating critical transitions in epithelial-hybrid-mesenchymal cell-fate determination”. In: Proceedings of the National Academy of Sci- ences of the United States of America 116.52 (2019), pp. 26343–26352. doi: 10 . 1073 / pnas . 1913773116. url: https :...

  39. [47]

    Optimal-Transport Analysis of Single-Cell Gene Expression Identifies Developmental Trajectories in Reprogramming

    G. Schiebinger et al. “Optimal-Transport Analysis of Single-Cell Gene Expression Identifies Developmental Trajectories in Reprogramming”. In: Cell 176.4 (2019), 928–943.e22. doi: 10.1016/j.cell.2019.01.006

  40. [48]

    Loss of KDM6A confers drug resistance in acute myeloid leukemia

    S. M. Stief et al. “Loss of KDM6A confers drug resistance in acute myeloid leukemia”. In: Leukemia 34.1 (2020), pp. 50–62. doi: 10.1038/s41375- 019-0497-6

  41. [49]

    Transformers in single-cell omics: a review and new perspectives

    Agata Sza lata, Klemen Hrovatin, Sarah Becker, et al. “Transformers in single-cell omics: a review and new perspectives”. In: Nature Methods 21 (2024), pp. 1430–1443. doi: 10.1038/s41592-024-02353-z

  42. [50]

    TrajectoryNet: A Dynamic Optimal Transport Network for Modeling Cellular Dynamics

    A. Tong et al. “TrajectoryNet: A Dynamic Optimal Transport Network for Modeling Cellular Dynamics”. In: Proc Mach Learn Res 119 (2020), pp. 9526–9536

  43. [51]

    Self-Organizing Global Gene Expression Regulated through Criticality: Mechanism of the Cell-Fate Change

    M. Tsuchiya et al. “Self-Organizing Global Gene Expression Regulated through Criticality: Mechanism of the Cell-Fate Change”. In: PLOS ONE 11.12 (2016), e0167912. doi: 10 . 1371 / journal . pone . 0167912. url: https://doi.org/10.1371/journal.pone.0167912

  44. [52]

    Transcriptome free energy can serve as a dynamic patient- specific biomarker in acute myeloid leukemia

    L. Uechi et al. “Transcriptome free energy can serve as a dynamic patient- specific biomarker in acute myeloid leukemia”. In: NPJ Systems Biology and Applications 10.1 (2024), p. 32. doi: 10.1038/s41540-024-00352-6

  45. [53]

    Free Energy Changes at State-Transition Critical Points As a Patient-Specific Biomarker in Acute Myeloid Leukemia

    Lisa Uechi et al. “Free Energy Changes at State-Transition Critical Points As a Patient-Specific Biomarker in Acute Myeloid Leukemia”. In: Blood 142.Supplement 1 (2023), p. 6012. doi: 10.1182/blood-2023-190662. 36

  46. [54]

    Cell Fate Dynamics Reconstruction Identifies TPT1 and PTPRZ1 Feedback Loops as Master Regulators of Differentiation in Pediatric Glioblastoma-Immune Cell Networks

    A. Uthamacumaran. “Cell Fate Dynamics Reconstruction Identifies TPT1 and PTPRZ1 Feedback Loops as Master Regulators of Differentiation in Pediatric Glioblastoma-Immune Cell Networks”. In: Interdisciplinary Sci- ences 17.1 (Mar. 2025). Epub 2024 Oct 17, pp. 59–85. doi: 10 . 100...

  47. [55]

    Algorithmic reconstruction of glioblas- toma network complexity

    A. Uthamacumaran and M. Craig. “Algorithmic reconstruction of glioblas- toma network complexity”. In: iScience 25.5 (Mar. 2022), p. 104179. doi: 10.1016/j.isci.2022.104179 . url: https://doi.org/10.1016/j. isci.2022.104179

  48. [56]

    A Review of Mathemat- ical and Computational Methods in Cancer Dynamics

    Abicumaran Uthamacumaran and Hector Zenil. “A Review of Mathemat- ical and Computational Methods in Cancer Dynamics”. In: Frontiers in Oncology 12 (2022), p. 850731. doi: 10.3389/fonc.2022.850731

  49. [57]

    Blood-derived gene-expression profiling in unravelling susceptibility to recessive disease

    P. Vahteristo et al. “Blood-derived gene-expression profiling in unravelling susceptibility to recessive disease”. In: Journal of Medical Genetics 44.11 (2007), pp. 718–720. doi: 10.1136/jmg.2007.052431

  50. [58]

    Structure-function relationship of ASH1L and histone H3K36 and H3K4 methylation

    K. R. Vann, R. Sharma, C. C. Hsu, et al. “Structure-function relationship of ASH1L and histone H3K36 and H3K4 methylation”. In: Nature Com- munications 16 (2025), p. 2235. doi: 10 .1038 /s41467 - 025 - 57556 - 5. url: https://doi.org/10.1038/s41467-025-57556-5

  51. [59]

    Inferred Developmental Origins of Brain Tumors from Single Cell RNA-Sequencing Data

    Su Wang et al. “Inferred Developmental Origins of Brain Tumors from Single Cell RNA-Sequencing Data”. In: Neuro-Oncology Advances (2025). doi: 10 . 1093 / noajnl / vdaf016. url: https : / / doi . org / 10 . 1093 / noajnl/vdaf016

  52. [60]

    Evaluating Uses of Deep Learning Methods for Causal Inference

    Albert Whata and Charles Chimedza. “Evaluating Uses of Deep Learning Methods for Causal Inference”. In: IEEE Access 10 (2022). Published January 4, 2022, pp. 4052–4067. doi: 10.1109/ACCESS.2021.3140189 . url: https://doi.org/10.1109/ACCESS.2021.3140189

  53. [62]

    STGRNS: an interpretable transformer-based method for inferring gene regulatory networks from single-cell transcriptomic data

    Jing Xu et al. “STGRNS: an interpretable transformer-based method for inferring gene regulatory networks from single-cell transcriptomic data”. In: Bioinformatics 39 (4 2023). doi: 10.1093/bioinformatics/btad165

  54. [63]

    The epigenome of AML stem and progenitor cells

    Jun Yamazaki et al. “The epigenome of AML stem and progenitor cells”. In: Epigenetics 8.1 (2013), pp. 92–104. doi: 10.4161/epi.23243

  55. [64]

    Dissecting Subtype-Specific Tumor-Time Interactions and Underlying Hidden Drivers in Pediatric Acute Myeloid Leukemia Via Single-Cell Multi-Omics

    Jiyuan Yang et al. “Dissecting Subtype-Specific Tumor-Time Interactions and Underlying Hidden Drivers in Pediatric Acute Myeloid Leukemia Via Single-Cell Multi-Omics”. In: Blood 142.Supplement 1 (2023), p. 5977. issn: 0006-4971. 37

  56. [65]

    Deep learning of gene relationships from single cell time-course expression data

    Y. Yuan and Z. Bar-Joseph. “Deep learning of gene relationships from single cell time-course expression data”. In: Brief Bioinform 22.5 (2021), bbab142. doi: 10.1093/bib/bbab142

  57. [66]

    Causal deconvolution by algorithmic generative models

    Hector Zenil, Narsis A Kiani, and Angel A Zea. “Causal deconvolution by algorithmic generative models”. In: Nature Machine Intelligence 1 (2019), pp. 58–66. doi: 10.1038/s42256-018-0005-0

  58. [67]

    A Decomposition Method for Global Evaluation of Shannon Entropy and Local Estimations of Algorithmic Complexity

    Hector Zenil et al. “A Decomposition Method for Global Evaluation of Shannon Entropy and Local Estimations of Algorithmic Complexity”. In: Entropy 20.8 (2018), p. 605. doi: 10.3390/e20080605

  59. [68]

    An Algorithmic Information Calculus for Causal Dis- covery and Reprogramming Systems

    Hector Zenil et al. “An Algorithmic Information Calculus for Causal Dis- covery and Reprogramming Systems”. In: iScience 19 (2019), pp. 1160–

  60. [70]

    AML as a Complex Adaptive Ecosystem: The presence of neurotransmit- ter signaling, immune–metabolic programs, lymphoid transcriptional features, and microenvironmental remodeling signals—including cytoskeletal and extra- cellular matrix cues, bioelectric signals, ion channels,...

  61. [71]

    In this framework, cancer arises from miscommunication across regulatory networks and loss of co- herent feedback loops

    AML as a Complex Cybernetic System: The fragmentation and dissoci- ation of cell identity, and malignant plasticity can be interpreted as emergent, 43 behavioral patterns of a maladaptive cybernetic response to disrupted informa- tion flow during pattern formation, and nice co...

  62. [72]

    AML Plasticity as a Cognitive Engine Steering Maladaptive Behaviors: Lastly, our findings underscore complexity not merely as a byproduct of patho- genesis, but as a defining hallmark of leukemogenesis, and cancer progression. Emergent behaviors, nonlinear dynamics, phase tran...

  63. [1172]

    tipping points

    doi: 10.1016/j.isci.2019.07.043. 38 A Phenotypic Plasticity and the Need for Pre- dictive Algorithms to Decode Cell Fate Cy- bernetics Pediatric acute myeloid leukemia (pAML) remains a lethal hematological ma- lignancy and an evolutionary disorder driven by maladaptive behavio...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.