Pith. sign in

REVIEW 4 major objections 5 minor 51 references

GPO-VAE: Modeling Explainable Gene Perturbation Responses utilizing GRN-Aligned Parameter Optimization

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read GPO-VAE claims that a single variational autoencoder can simultaneously predict transcriptional responses to gene perturbations and recover an interpretable gene regulatory network from the Bernoulli parameter matrix $W$ that generates…

desk verdict A novel VAE with a GRN-shaped mask gives solid perturbation predictions, but the GRN explainability claim is undermined by a signed-expression representation gap and circular evaluation. read the letter →

arxiv 2501.18973 v1 pith:ARDDPWWR submitted 2025-01-31 cs.LG cs.AI

Seungheun Baek , Soyon Park , Yan Ting Chok , Mogan Gim , Jaewoo Kang This is my paper · ORCID
classification cs.LGcs.AI
keywords geneperturbationresponseregulatorynetworkexplainablevariationalautoencodersparsemechanismshiftPerturb-seqcausalgraphinferencedifferentialexpressionK-hopaccumulation
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

GPO-VAE is a variational autoencoder for single-cell gene perturbation data that tries to make its latent perturbation effect explicitly interpretable as a gene regulatory network (GRN). The paper's claim is that the same model can predict post-perturbation transcriptional responses at the top of its baseline comparisons and, from the very matrix $W$ that controls its sparse latent perturbation masks, recover a directed gene-to-gene causal graph that agrees with known biology, including KRAS, MYC, and NTRK1 subnetworks, and with statistical tests of causal effect. The mechanism is a GRN-aligned parameter optimization: a differential-expression loss pushes $W$ to reproduce observed fold changes, a $K$-hop accumulation term lets extended genes enter through powers of $W$, and an $ℓ^1$ penalty keeps the graph sparse. If correct, this dissolves the usual trade-off between prediction performance and explainability in perturbation modeling, because the explainable structure is not a post-hoc analysis but the parameterization of the predictor itself.

What carries the argument

The load-bearing object is the square Bernoulli parameter matrix $W \in [0,1]^{(|G^\circ \cup G^+|) \times (|G^\circ \cup G^+|)}$, reinterpreted as a weighted adjacency matrix of gene-to-gene causal probabilities with diagonal entries as self-regulation. It is trained by the GPO objective $J_{\mathrm{gpo}}(W)=\|P T_K - \Delta X\|_1 + \|W\|_1$, where $\Delta X$ is an optimal-transport-paired differential expression reference, $T_K = W + \sum_{k=2}^K \frac{1}{|G^\circ \cup G^+|} W^k$ accumulates multi-hop causal paths with $K=5$, and the $ℓ^1$ term enforces sparsity. The same $W$ parameterizes the Bernoulli sampler of the latent perturbation encoder, so the GRN is not a separate inference module but the very mechanism by which the VAE generates perturbation-specific latent offsets, and it is thresholded at 0.5 to produce the reported graph.

What would settle it

A direct check: take a known repressor perturbation whose knockdown raises a target's expression; if the model must keep $W$ in $[0,1]$ and the loss only matches $P W$ to the negative differential expression, the predicted effect on that target cannot go negative, so the $W$-as-GRN interpretation fails whenever such repression is observed.

Watch

Extended reading notes

Core claim

The central discovery is that redesigning the latent perturbation encoder so its Bernoulli mask probabilities form a square matrix $W$ over perturbed and extended genes, and then optimizing $W$ with the GPO objective $J_{\mathrm{gpo}}(W)=\|P T_K - \Delta X\|_1 + \|W\|_1$ where $T_K = W + \sum_{k=2}^K \frac{1}{|G^\circ \cup G^+|} W^k$, yields both stronger perturbation-response prediction than prior VAE baselines on the Replogle K562, Replogle RPE1, and Adamson datasets, and a sparse thresholded GRN with lower false omission rate and higher mean Wasserstein distance than dedicated causal-discovery baselines. The paper reads each $W_{i,j}$ as the causal probability from gene $i$ to gene $j$, treats $W^k$ as accumulated $k$-hop causal relations, and validates the recovered subnetworks against experimentally supported KRAS, MYC, and NTRK1 pathways. The claim is that one learned object carries both jobs: sampling masks for the VAE and forming a biologically meaningful adjacency graph.

Load-bearing premise

The load-bearing premise is that each entry of the learned matrix $W$, kept between 0 and 1 and matched to signed differential expression through an $ℓ^1$ loss, is a genuine causal gene-to-gene probability, even though a downregulated interaction would have to be encoded as $W$ near 0 and is then indistinguishable from 'no edge' after the 0.5 threshold.

Editorial extensions

If this is right

  • If $W$ is a true GRN, every predicted perturbation response is traceable: the change for a target gene is explained by the edges and multi-hop paths through which the perturbed gene influences it.
  • Because $T_K$ includes paths through extended genes, genes that were never experimentally perturbed can enter explanations, and the model can generalize to unseen perturbation treatments; the paper demonstrates this on held-out TWISTNB, RPL26, and RPL34 perturbations.
  • The sparse graph recovered from $W$ achieves better mean Wasserstein distance and false omission rate than dedicated GRN inference baselines, so network inference and perturbation prediction are accomplished in a single training run.
  • Biologically, subnetworks that pass through extended genes nominate candidate interactions, such as KRAS synthetic-lethal partners and MYC- or NTRK1-pathway genes, that curated interaction databases do not contain, yielding concrete testable hypotheses.
  • The model's architecture does not restrict it to single-gene perturbations, so the same GPO objective could in principle be extended to multi-gene treatments and to synergy or inhibition relationships, although the paper does not implement that extension.

Reading between the lines

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

  • Editorial inference: the paper never shows how signed, negative differential expression is encoded by $W \in [0,1]$, so a direct test of the causal reading is to perturb a known repressor and check whether predicted expression of its targets moves below control levels.
  • Editorial inference: the term $W^k$ counts walks that may revisit nodes, so the 'k-hop causal relationship' interpretation presumes cycles are meaningful; a DAG-constrained or self-loop-free variant of the GPO loss would clarify whether the multi-hop term accumulates genuine causal paths or mostly adds correlation.
  • Editorial inference: because the differential expression reference is built from the same training data that measures GRN quality, the reported µWD and FOR improvements could partly reflect fitting to the evaluation signal; a stronger test would be to infer edges for a cell type or perturbation set never used in training and compare against a known reference network.
  • Editorial inference: if the causal reading of $W$ is correct, the method offers a template for other structured latent spaces where the VAE's sampling distribution is itself the biological quantity of interest, but the probability interpretation needs independent identifiability or external validation before that template can be relied on.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes GPO-VAE, a variational autoencoder for single-cell gene perturbation response prediction in which the latent perturbation effect is governed by a square Bernoulli parameter matrix W that is interpreted as a gene regulatory network (GRN). A GRN-aligned parameter optimization (GPO) loss (Eq. 20) fits W, through a multi-hop accumulated matrix T_K and an L1 sparsity penalty, to the signed differential expression between perturbed and control samples. Experiments on Replogle K562, Replogle RPE1, and Adamson report state-of-the-art ATE correlation/R²/Jaccard among VAE-based baselines, and the GRN obtained by thresholding W at 0.5 is reported to achieve the best µWD and FOR against CausalBench baselines. Qualitative case studies discuss KRAS-, MYC-, and NTRK1-related subnetworks, and a case study reports competitive performance on three held-out perturbation treatments.

Significance. If the central claim holds, the contribution is valuable: a single VAE would simultaneously achieve strong perturbation-response prediction and yield an interpretable, data-consistent regulatory network. The paper ships public code, reports results over multiple seeds with standard deviations, and includes a systematic ablation (Table 5) showing that each GPO loss component contributes. The perturbation-prediction comparisons in Table 3 are internally consistent and support the state-of-the-art claim within the VAE family on these datasets. However, the explainability claim rests on reading W as a matrix of causal probabilities in [0,1], while the GPO loss fits signed differential expression. Because P T_K is nonnegative under the stated parameterization, the model cannot represent downregulation except by pushing W toward zero, which is indistinguishable from 'no edge' after thresholding. This is a load-bearing correctness risk for the GRN-inference and biological-interpretation claims, and it also casts doubt on the reported µWD/FOR improvements.

major comments (4)
  1. [§2.2.5, Eq. (15)-(20)] W is described as a matrix of causal gene-to-gene probabilities with entries in [0,1] (also stated in §3.2), and Eq. (1) samples Bernoulli masks from W. Yet no constraint or reparameterization (e.g., sigmoid) is specified, and the GPO loss J_gpo = Σ||P T_K − ΔX||_1 + ||W||_1 fits the signed differential expression matrix ΔX directly. Since P T_K is nonnegative under Eq. (18), a downregulated target can only be fitted by driving W toward zero; after thresholding at 0.5 this is indistinguishable from 'no edge'. The inferred GRN therefore cannot represent repression, which is a core feature of gene regulation. Please reparameterize W with signed weights (or separate positive/negative edge matrices) and retrain, or otherwise show how the current [0,1] parameterization can represent signed fold changes; at a minimum, report the distribution of signs in ΔX and the model's ability to capture them.
  2. [§2.2.5, Eq. (18)] The assertion that W^k, normalized by 1/|G°∪G+|, 'contains k-hop causal relationships' is stated without derivation. For a matrix with entries in [0,1], powers are not Bernoulli probability matrices and row sums are not preserved, so the normalization is ad hoc; moreover, powers of a nonnegative matrix only accumulate positive regulatory paths, compounding the sign bias identified above. The paper should either provide a derivation of this multi-hop interpretation or validate it empirically, for example by showing that the k-hop terms improve prediction of both upregulated and downregulated targets rather than only positive effects.
  3. [§3.2 and Table 4] The GRN evaluation metrics µWD and FOR are computed from perturbation-versus-control expression changes, which include strongly downregulated genes. Since the current W cannot represent negative edges, edges with strong negative target shifts cannot be selected, so the reported 'best µWD and FOR' for GPO-VAE in Table 4 may partly reflect this representational bias rather than genuine recovery of causal regulatory structure. The paper should report the signed edge analysis and, ideally, recompute µWD separately for positive- and negative-effect target genes under a signed reparameterization.
  4. [§2.2.5 vs. §3.2] The GRN inference result is partly circular: the GPO loss in Eq. (20) explicitly minimizes ||P T_K − ΔX||_1, where ΔX is the differential expression between perturbed and control samples, and the CausalBench-style evaluation in §3.2 scores an edge A→B by the perturbation-vs-control distribution shift of B when A is perturbed. The model is therefore being scored on a quantity closely tied to the loss it was trained to minimize, unlike the unsupervised causal-discovery baselines. The paper should acknowledge this and provide a nontrivial test of the GRN claim, such as held-out perturbation evaluation, comparison against a supervised baseline fitted to ΔX, or evaluation on independent pathway databases with proper significance testing.
minor comments (5)
  1. [§2.2.5] K is introduced only after Eq. (18) ('K is equally set to 5'); please define K before its first use in Eq. (17).
  2. [§3.2 and Figure 3] The edge threshold is stated as 0.5, but Figure 3 distinguishes 'initialized parameter (=0.5)' from 'presence of edge (>0.5)'. Please clarify whether the threshold is strict (>0.5) or non-strict (≥0.5), since this changes the number of edges in Table 4.
  3. [Table 4] The Random 100,000 baseline row reports FOR = -1.000 for all datasets; under the definition of FOR as a proportion of negative edges, negative values are not possible. Please check the baseline computation or the metric definition.
  4. [Table 5] The text says the full GPO-VAE gives a 'slight improvement in perturbation response prediction' over its ablations, but on RPE1 the full model has ATE-ρ 0.6584, which is slightly lower than the J_K_dge ablation's 0.6593. Please qualify the statement or note that the improvement is not uniform across metrics.
  5. [§3.5] The pathway case studies are suggestive, but no enrichment p-values or multiple-testing corrections are reported. Please add adjusted enrichment statistics or explicitly describe these analyses as anecdotal.

Circularity Check

1 steps flagged · score 7.0 of 10

The GRN 'inference' result is largely a self-consistency score: W is fit to the differential-expression target ΔX (Eq. 20), and the µWD/FOR metrics are computed from the same control-versus-perturbed shifts, so the top GRN scores are forced by construction rather than independent causal recovery.

  1. fitted input called prediction [§2.2.5, Eqs. (15)–(20); §3.2, µWD/FOR definitions]
    "we exploit the post-perturbational differential gene expression (DGE) values, which are calculated based on fold changes between control and perturbation gene expression profiles, as reference in adjusting the parameters. ... Jgpo( ˆW ) = J K dge + Jsp (19) = X ||P ˆTK − ∆X||1 + || ˆW ||1 (20) ... µWD measures the average strength of causal effects of the inferred edges. For edge from A to B, WD is computed between the empirical distribution of the expression of B in control samples and in A-perturbed samples."

    W is optimized by minimizing ||P T_K − ΔX||_1, where ΔX is the paired control-vs-perturbed differential expression (Eqs. 15–20). The GRN metrics then score edge A→B by the same control-vs-perturbed shift: µWD is the Wasserstein distance between B in control and in A-perturbed cells, and FOR is a Mann–Whitney test on those same distributions. Edges with high µWD are therefore precisely the A→B effects the loss was trained to fit in ΔX, so the reported 'best µWD and FOR' are self-consistency scores for the training target, not independent causal recovery. Since W∈[0,1] while ΔX is signed, downregulation is forced toward zero and read as 'no edge' at the 0.5 threshold, further biasing the comparison toward activating edges.

full rationale

The perturbation-response results (Table 3) are held-out predictions and are not circular in themselves: the GPO loss is an auxiliary supervised term, but the ATE/Jaccard metrics are computed on test perturbations, so that part has independent content. The circularity is concentrated in the GRN-inference evaluation (Table 4): the extracted W is the very parameter fit to ΔX, and the µWD/FOR metrics are computed from the same control-versus-perturbed expression shifts that define ΔX. The paper's own conclusion also concedes 'the absence of ground truth limits the comprehensiveness of existing evaluation methods,' which matches this reading. No load-bearing self-citation chain is needed for this finding; citations to the authors' CRADLE-VAE are architectural/preprocessing inheritances, not the source of the GRN claim. Score 7 reflects that the central explainability claim reduces largely to a fit, while the response-prediction half retains independent benchmark content.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The load-bearing content of GPO-VAE is: (1) two ad hoc semantic assumptions that turn W into a GRN and W^k into a multi-hop cause accumulator, (2) imported domain assumptions (Sparse Mechanism Shift, CausalBench's causal reading of perturbation effects, validity of OT pairing), and (3) hand-set hyperparameters K=5, β with per-dataset penalty coefficients, and the 0.5 edge threshold. The DGE loss itself drives both the ATE gains and the GRN results: it makes W a fitted surrogate of observed differential expression, which is precisely what the GRN evaluation metrics re-measure. No new biological entities are introduced; the inferred GRN is a learned construct whose only external evidence is the post-hoc pathway overlap.

free parameters (4)
  • K (hop number in J^K_dge) = 5
    Set equally to 5 for all datasets (§2.2.5); no sensitivity analysis reported, and it controls how much extended-gene signal enters the loss.
  • GPO loss weight (β and per-dataset penalty coefficient) = β=0.1; penalty coefficient 5 (K562), 10 (RPE1), 1 (Adamson)
    Appendix F reports different penalty coefficients per dataset; the mapping to Eq. 21's β is unclear, and these were hand-tuned.
  • GRN edge threshold = 0.5
    Edges are declared when W_ij > 0.5 (§3.2); hand-chosen, with no robustness check across thresholds.
  • Mask prior probability = 0.3
    Prior for Bernoulli mask sampling (Appendix F); inherited from CRADLE-VAE and hand-set, load-bearing for the sparsity of perturbation effects.
assumptions (6)
  • domain assumption Sparse Mechanism Shift hypothesis: only a few causal mechanisms change per perturbation
    Invoked in Section 1 as the design basis for the sparse perturbation mask; assumed true for these datasets.
  • ad hoc to paper W_ij is a causal probability from gene i to gene j, so the square parameter matrix is a GRN
    Section 2.2.5 redefines the Bernoulli parameters as causal probabilities; this semantic leap is enforced by a fitting loss rather than justified by an identification argument.
  • ad hoc to paper W^k, normalized by 1/|G°∪G+|, represents accumulated k-hop causal relationships
    Section 2.2.5, Eq. 18: asserted without derivation; powers of an unsigned probability matrix do not generally equal path probabilities.
  • domain assumption Optimal transport pairing of perturbed and control cells produces a valid differential expression reference ΔX
    Section 2.2.5 and B.2: the pairing is a modeling choice; sensitivity of the results to the pairing details is not studied.
  • domain assumption Perturbation-induced expression change of target B when A is perturbed measures the causal effect A to B
    Section 3.2 and D.3-D.4: µWD and FOR assume no confounding and that measurable expression shift equals causal regulation; the authors note the biological/statistical mismatch in Section 4.
  • standard math Standard variational inference, correlated variational family, and Gamma-Poisson decoding
    Sections 2.2.2-2.2.3: conventional ELBO and VAE machinery, unproblematic.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GPO-VAE: Modeling Explainable Gene Perturbation Responses utilizing GRN-Aligned Parameter Optimization." pith.science (2026). https://pith.science/paper/ARDDPWWR

@misc{pith2026250118973,
  author       = {Pith},
  title        = {Pith review of: GPO-VAE: Modeling Explainable Gene Perturbation Responses utilizing GRN-Aligned Parameter Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ARDDPWWR}},
  note         = {Machine review of arXiv:2501.18973}
}
read the original abstract

Motivation: Predicting cellular responses to genetic perturbations is essential for understanding biological systems and developing targeted therapeutic strategies. While variational autoencoders (VAEs) have shown promise in modeling perturbation responses, their limited explainability poses a significant challenge, as the learned features often lack clear biological meaning. Nevertheless, model explainability is one of the most important aspects in the realm of biological AI. One of the most effective ways to achieve explainability is incorporating the concept of gene regulatory networks (GRNs) in designing deep learning models such as VAEs. GRNs elicit the underlying causal relationships between genes and are capable of explaining the transcriptional responses caused by genetic perturbation treatments. Results: We propose GPO-VAE, an explainable VAE enhanced by GRN-aligned Parameter Optimization that explicitly models gene regulatory networks in the latent space. Our key approach is to optimize the learnable parameters related to latent perturbation effects towards GRN-aligned explainability. Experimental results on perturbation prediction show our model achieves state-of-the-art performance in predicting transcriptional responses across multiple benchmark datasets. Furthermore, additional results on evaluating the GRN inference task reveal our model's ability to generate meaningful GRNs compared to other methods. According to qualitative analysis, GPO-VAE posseses the ability to construct biologically explainable GRNs that align with experimentally validated regulatory pathways. GPO-VAE is available at https://github.com/dmis-lab/GPO-VAE

Figures

Figures reproduced from arXiv: 2501.18973 by the authors.

Figure 1
Figure 1. Model Overview of GPO-VAE. The model consists of three encoder modules: latent perturbation encoder, latent artifact encoder, and latent basal state encoder, and a decoder. The perturbation encoder utilizes a gene regulatory network. X ∈ R N×|G| , P ∈ {0, 1} N×|G◦∪G+| , A ∈ {0, 1} N×1 correspond to the matrices of gene expression profiles, one-hot encoded gene perturbation treatments and quality control (QC) labels … view at source ↗
Figure 2
Figure 2. Comparison of perturbation encoder between previous models and our model. Unlike previous models [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. GRN topology analysis of GPO loss objective. Red borderlines separate perturbation and extended gene [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Case study and pathway analysis of GRN subnetworks involved in the interaction with three cancer proteins: [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Quantitative performance of the model on unseen perturbations for TWISTNB, RPL26, and RPL34 genes. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: GRN topology analysis of GPO loss objective. Red borderlines separate perturbation and extended gene [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 47 canonical work pages

  1. [1]

    & Others Combinatorial single-cell CRISPR screens by direct guide RNA capture and targeted sequencing

    Replogle, J., Norman, T., Xu, A., Hussmann, J., Chen, J., Cogan, J., Meer, E., Terry, J., Riordan, D., Srinivas, N. & Others Combinatorial single-cell CRISPR screens by direct guide RNA capture and targeted sequencing. Nature Biotechnology. 38, 954-961 (2020)

    work page 2020

  2. [2]

    & Others A multiplexed single-cell CRISPR screening platform enables systematic dissection of the unfolded protein response

    Adamson, B., Norman, T., Jost, M., Cho, M., Nuñez, J., Chen, Y ., Villalta, J., Gilbert, L., Horlbeck, M., Hein, M. & Others A multiplexed single-cell CRISPR screening platform enables systematic dissection of the unfolded protein response. Cell. 167, 1867-1882 (2016)

    work page 2016

  3. [3]

    & Regev, A

    Rood, J., Hupalowska, A. & Regev, A. Toward a foundation model of causal cell and tissue biology with a Perturbation Cell and Tissue Atlas. Cell. 187, 4520-4545 (2024)

    work page 2024

  4. [4]

    & Schwab, P

    Chevalley, M., Roohani, Y ., Mehrjou, A., Leskovec, J. & Schwab, P. Causalbench: A large-scale benchmark for network inference from single-cell perturbation data. ArXiv Preprint arXiv:2210.17283. (2022)

    arXiv 2022

  5. [5]

    & Kang, J

    Baek, S., Park, S., Chok, Y ., Lee, J., Park, J., Gim, M. & Kang, J. CRADLE-V AE: Enhancing Single-Cell Gene Perturbation Modeling with Counterfactual Reasoning-based Artifact Disentanglement. ArXiv Preprint arXiv:2409.05484. (2024)

    arXiv 2024

  6. [6]

    & Psomopoulos, F

    Gavriilidis, G., Vasileiou, V ., Orfanou, A., Ishaque, N. & Psomopoulos, F. A mini-review on perturbation modelling across single-cell omic modalities. Computational And Structural Biotechnology Journal . (2024)

    work page 2024

  7. [7]

    & Morris, S

    Kamimoto, K., Stringa, B., Hoffmann, C., Jindal, K., Solnica-Krezel, L. & Morris, S. Dissecting cell identity via network inference and in silico gene perturbation. Nature. 614, 742-751 (2023)

    work page 2023

  8. [8]

    & Karaletsos, T

    Bereket, M. & Karaletsos, T. Modelling cellular perturbations with the sparse additive mechanism shift variational autoencoder. Advances In Neural Information Processing Systems . 36 (2024)

    work page 2024

Show all 51 references
  1. [9]

    & Gong, M

    Xu, D., Gao, E., Huang, W., Wang, M., Song, A. & Gong, M. On the sparse DAG structure learning based on adaptive Lasso. ArXiv Preprint arXiv:2209.02946. (2022)

    2022 arXiv

  2. [10]

    & Regev, A

    Lopez, R., Tagasovska, N., Ra, S., Cho, K., Pritchard, J. & Regev, A. Learning causal representations of single cells via sparse mechanism shift modeling. Conference On Causal Learning And Reasoning . pp. 662-691 (2023)

    2023

  3. [11]

    & Yan, X

    Sohn, K., Lee, H. & Yan, X. Learning structured output representation using deep conditional generative models. Advances In Neural Information Processing Systems . 28 (2015)

    2015

  4. [12]

    & Scheines, R

    Spirtes, P., Glymour, C. & Scheines, R. Causation, prediction, and search. (MIT press,2001)

    2001

  5. [13]

    Optimal structure identification with greedy search

    Chickering, D. Optimal structure identification with greedy search. Journal Of Machine Learning Research. 3, 507-554 (2002)

    2002

  6. [14]

    & Xing, E

    Zheng, X., Aragam, B., Ravikumar, P. & Xing, E. Dags with no tears: Continuous optimization for structure learning. Advances In Neural Information Processing Systems . 31 (2018)

    2018

  7. [15]

    & Weichwald, S

    Reisach, A., Seiler, C. & Weichwald, S. Beware of the simulated dag! causal discovery benchmarks may be easy to game. Advances In Neural Information Processing Systems . 34 pp. 27772-27784 (2021)

    2021

  8. [16]

    & Others SCENIC: single-cell regulatory network inference and clustering.Nature Methods

    Aibar, S., González-Blas, C., Moerman, T., Huynh-Thu, V ., Imrichova, H., Hulselmans, G., Rambow, F., Marine, J., Geurts, P., Aerts, J. & Others SCENIC: single-cell regulatory network inference and clustering.Nature Methods. 14, 1083-1086 (2017)

    2017

  9. [17]

    & Bühlmann, P

    Hauser, A. & Bühlmann, P. Characterization and greedy learning of interventional Markov equivalence classes of directed acyclic graphs. The Journal Of Machine Learning Research . 13, 2409-2464 (2012) 12 A PREPRINT - FEBRUARY 3, 2025

    2012

  10. [18]

    & Drouin, A

    Brouillard, P., Lachapelle, S., Lacoste, A., Lacoste-Julien, S. & Drouin, A. Differentiable causal discovery from interventional data. Advances In Neural Information Processing Systems . 33 pp. 21865-21877 (2020)

    2020

  11. [19]

    & Regev, A

    Lopez, R., Hütter, J., Pritchard, J. & Regev, A. Large-scale differentiable causal discovery of factor graphs. Advances In Neural Information Processing Systems . 35 pp. 19290-19303 (2022)

    2022

  12. [20]

    & Others Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq

    Replogle, J., Saunders, R., Pogson, A., Hussmann, J., Lenail, A., Guna, A., Mascibroda, L., Wagner, E., Adelman, K., Lithwick-Yanai, G. & Others Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq. Cell. 185, 2559-2575 (2022)

    2022

  13. [21]

    & Others The STRING database in 2023: protein–protein association networks and functional enrichment analyses for any sequenced genome of interest

    Szklarczyk, D., Kirsch, R., Koutrouli, M., Nastou, K., Mehryary, F., Hachilif, R., Gable, A., Fang, T., Doncheva, N., Pyysalo, S. & Others The STRING database in 2023: protein–protein association networks and functional enrichment analyses for any sequenced genome of interest....

    2023

  14. [22]

    & Ruepp, A

    Tsitsiridis, G., Steinkamp, R., Giurgiu, M., Brauner, B., Fobo, G., Frishman, G., Montrone, C. & Ruepp, A. CORUM: the comprehensive resource of mammalian protein complexes–2022. Nucleic Acids Research . 51, D539-D545 (2023)

    2023

  15. [23]

    Liu, Z., Wu, C., Miao, H. & Wu, H. RegNetwork: an integrated database of transcriptional and post-transcriptional regulatory networks in human and mouse. Database. 2015 pp. bav095 (2015)

    2015

  16. [24]

    & Others The BioGRID database: A comprehensive biomedical resource of curated protein, genetic, and chemical interactions

    Oughtred, R., Rust, J., Chang, C., Breitkreutz, B., Stark, C., Willems, A., Boucher, L., Leung, G., Kolas, N., Zhang, F. & Others The BioGRID database: A comprehensive biomedical resource of curated protein, genetic, and chemical interactions. Protein Science. 30, 187-200 (2021)

    2021

  17. [25]

    & Poole, B

    Jang, E., Gu, S. & Poole, B. Categorical reparameterization with gumbel-softmax. ArXiv Preprint arXiv:1611.01144. (2016)

    2016 arXiv

  18. [26]

    & Hartford, J

    Atanackovic, L., Tong, A., Wang, B., Lee, L., Bengio, Y . & Hartford, J. Dyngfn: Towards Bayesian inference of gene regulatory networks with gflownets.Advances In Neural Information Processing Systems. 36 pp. 74410-74428 (2023)

    2023

  19. [27]

    & Bengio, Y

    Schölkopf, B., Locatello, F., Bauer, S., Ke, N., Kalchbrenner, N., Goyal, A. & Bengio, Y . Toward causal represen- tation learning. Proceedings Of The IEEE. 109, 612-634 (2021)

    2021

  20. [28]

    & Herrera, F

    Charte, D., Charte, F., Jesus, M. & Herrera, F. An analysis on the use of autoencoders for representation learning: Fundamentals, learning task case studies, explainability and challenges. Neurocomputing. 404 pp. 93-107 (2020)

    2020

  21. [29]

    & Consortium, P

    Amann, J., Blasimme, A., Vayena, E., Frey, D., Madai, V . & Consortium, P. Explainability for artificial intelligence in healthcare: a multidisciplinary perspective. BMC Medical Informatics And Decision Making . 20 pp. 1-9 (2020)

    2020

  22. [30]

    & Zemel, R

    Klys, J., Snell, J. & Zemel, R. Learning latent subspaces in variational autoencoders. Advances In Neural Information Processing Systems. 31 (2018)

    2018

  23. [31]

    & Tenenbaum, J

    Kulkarni, T., Whitney, W., Kohli, P. & Tenenbaum, J. Deep convolutional inverse graphics network.Advances In Neural Information Processing Systems. 28 (2015)

    2015

  24. [32]

    Li, P. & Wu, M. Learning to refine domain knowledge for biological network inference. ArXiv Preprint arXiv:2410.14436. (2024)

    2024 arXiv

  25. [33]

    & Others WikiPathways 2024: next generation pathway database

    Agrawal, A., Balcı, H., Hanspers, K., Coort, S., Martens, M., Slenter, D., Ehrhart, F., Digles, D., Waagmeester, A., Wassink, I. & Others WikiPathways 2024: next generation pathway database. Nucleic Acids Research. 52, D679-D689 (2024)

    2024

  26. [34]

    (2024), https://thebiogrid.org/121697/summary/homo-sapiens/plekha4.html, Accessed: 2024-01-20

    BioGRID PLEKHA4. (2024), https://thebiogrid.org/121697/summary/homo-sapiens/plekha4.html, Accessed: 2024-01-20

    2024

  27. [35]

    & Baskin, J

    Shah, A., Batrouni, A., Kim, D., Punyala, A., Cao, W., Han, C., Goldberg, M., Smolka, M. & Baskin, J. PLEKHA4/kramer attenuates dishevelled ubiquitination to modulate Wnt and planar cell polarity signaling. Cell Reports. 27, 2157-2170 (2019)

    2019

  28. [36]

    (2024), https://www.wikipathways.org/pathways/WP3888.html, Accessed: 2024-01-20

    WikiPathways WP3888: Wnt Signaling Pathway. (2024), https://www.wikipathways.org/pathways/WP3888.html, Accessed: 2024-01-20

    2024

  29. [37]

    & Chang, W

    Sherman, B., Hao, M., Qiu, J., Jiao, X., Baseler, M., Lane, H., Imamichi, T. & Chang, W. DA VID: a web server for functional enrichment analysis and functional annotation of gene lists (2021 update). Nucleic Acids Research. 50, W216-W221 (2022)

    2022

  30. [38]

    & Others Integrative oncogene-dependency mapping identifies RIT1 vulnerabilities and synergies in lung cancer

    Vichas, A., Riley, A., Nkinsi, N., Kamlapurkar, S., Parrish, P., Lo, A., Duke, F., Chen, J., Fung, I., Watson, J. & Others Integrative oncogene-dependency mapping identifies RIT1 vulnerabilities and synergies in lung cancer. Nature Communications. 12, 4789 (2021) 13 A PREPRINT...

    2021

  31. [39]

    & Others MYC multimers shield stalled replication forks from RNA polymerase

    Solvie, D., Baluapuri, A., Uhl, L., Fleischhauer, D., Endres, T., Papadopoulos, D., Aziba, A., Gaballa, A., Mikicic, I., Isaakova, E. & Others MYC multimers shield stalled replication forks from RNA polymerase. Nature. 612, 148-155 (2022)

    2022

  32. [40]

    & Qing, G

    Wang, L., Chen, C., Song, Z., Wang, H., Ye, M., Wang, D., Kang, W., Liu, H. & Qing, G. EZH2 depletion potentiates MYC degradation inhibiting neuroblastoma and small cell carcinoma tumor formation. Nature Communications. 13, 12 (2022)

    2022

  33. [41]

    & Beli, P

    Heidelberger, J., V oigt, A., Borisova, M., Petrosino, G., Ruf, S., Wagner, S. & Beli, P. Proteomic profiling of VCP substrates links VCP to K6-linked ubiquitylation and c-Myc function. EMBO Reports. 19, e44754 (2018)

    2018

  34. [42]

    & Jensen, L

    Binder, J., Pletscher-Frankild, S., Tsafou, K., Stolte, C., O’Donoghue, S., Schneider, R. & Jensen, L. COMPART- MENTS: unification and visualization of protein subcellular localization evidence. Database. 2014 (2014)

    2014

  35. [43]

    & Others UniProt: the Universal Protein Knowledgebase in 2025

    Bateman, A., Martin, M., Orchard, S., Magrane, M., Adesina, A., Ahmad, S., Bowler-Barnett, E., Bye-A-Jee, H., Carpentier, D., Denny, P. & Others UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Research. (2024)

    2024

  36. [44]

    & Hockenbery, D

    Morrish, F. & Hockenbery, D. MYC and mitochondrial biogenesis. Cold Spring Harbor Perspectives In Medicine . 4, a014225 (2014)

    2014

  37. [45]

    & Tognon, C

    Joshi, S., Davare, M., Druker, B. & Tognon, C. Revisiting NTRKs as an emerging oncogene in hematological malignancies. Leukemia. 33, 2563-2574 (2019)

    2019

  38. [46]

    & Others Somatic MED12 mutations are associated with poor prognosis markers in chronic lymphocytic leukemia

    Kämpjärvi, K., Järvinen, T., Heikkinen, T., Ruppert, A., Senter, L., Hoag, K., Dufva, O., Kontro, M., Rassenti, L., Hertlein, E. & Others Somatic MED12 mutations are associated with poor prognosis markers in chronic lymphocytic leukemia. Oncotarget. 6, 1884 (2014)

    2014

  39. [47]

    & Others Effective tumor cell abrogation via Venetoclax-mediated BCL-2 inhibition in KMT2A-rearranged acute B-lymphoblastic leukemia

    Richter, A., Lange, S., Holz, C., Brock, L., Freitag, T., Sekora, A., Knuebel, G., Krohn, S., Schwarz, R., Hinz, B. & Others Effective tumor cell abrogation via Venetoclax-mediated BCL-2 inhibition in KMT2A-rearranged acute B-lymphoblastic leukemia. Cell Death Discovery. 8, 302 (2022)

    2022

  40. [48]

    & Others Somatic TP53 mutations characterize preleukemic stem cells in acute myeloid leukemia

    Lal, R., Lind, K., Heitzer, E., Ulz, P., Aubell, K., Kashofer, K., Middeke, J., Thiede, C., Schulz, E., Rosenberger, A. & Others Somatic TP53 mutations characterize preleukemic stem cells in acute myeloid leukemia. Blood, The Journal Of The American Society Of Hematology . 129...

    2017

  41. [49]

    & Diercks, A

    Nazitto, R., Amon, L., Mast, F., Aitchison, J., Aderem, A., Johnson, J. & Diercks, A. ILF3 is a negative transcriptional regulator of innate immune responses and myeloid dendritic cell maturation. The Journal Of Immunology. 206, 2949-2965 (2021)

    2021

  42. [50]

    & Others Optimal transport: old and new

    Villani, C. & Others Optimal transport: old and new. (Springer,2009) 14 A PREPRINT - FEBRUARY 3, 2025 A Quality Control (QC) Criteria We adopted the six QC criteria defined in CRADLE-V AE [5], in line with 10X Genomics, for the QC annotation of each of the gene expression data...

    2009

  43. [51]

    What would the model predict if a different action had been taken?

    Mitochondrial Read Percentage: The proportion of RNA reads originating from mitochondrial genes is an indicator of cellular health. Cells with a high percentage of mitochondrial RNA are often stressed or damaged, making this metric a valuable filter during quality control. 4. ...

    2025

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.