Pith. sign in

REVIEW 3 major objections 4 minor 300 references

The thesis argues that gene regulatory network inference is best split into two stages: a sequence-based prior that transfers across species and a probabilistic factorization that refines edges with calibrated uncertainty.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 21:28 UTC pith:TEMT24EL

load-bearing objection A solid two-method thesis whose main real weakness is that the headline transferability claim rests on cross-species experiments that don't report homology-based leakage controls. the 3 major comments →

arxiv 2607.16053 v1 pith:TEMT24EL submitted 2026-07-17 stat.ML cs.LGstat.APstat.ME

Deep and Probabilistic Models for Gene Regulatory Network Inference

Claudia Skok Gibbs This is my paper
classification stat.ML cs.LGstat.APstat.ME
keywords gene regulatory networksprobabilistic matrix factorizationvariational inferencesingle-cell RNA-seqsequence language modelstransfer learninguncertainty calibrationtranscription factors
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This thesis tries to establish that the two persistent failures of gene regulatory network inference—heuristic, confidence-free expression modeling and assay-dependent, species-specific priors—have separate fixes that work best in sequence. It presents PMF-GRN, a probabilistic matrix factorization of single-cell expression in which transcription-factor activity and TF-target interactions are latent, optimized by variational inference, so model selection is principled and every edge comes with a posterior variance that is demonstrably calibrated. It then presents GLM-Prior, a genomic language model fine-tuned to score TF-gene interaction from nucleotide sequence alone, which transfers across yeast, mouse, and human settings. Together they support a dual-stage view: a transferable sequence-derived scaffold first, then expression-based refinement with quantified uncertainty. If the thesis is right, GRN inference becomes usable in organisms and cell types where chromatin or ChIP-seq priors do not exist, and predictions come with a rank ordering of confidence for experimental follow-up.

Core claim

On the thesis's own terms, the central discovery is that GRN reconstruction decomposes cleanly into two complementary problems. PMF-GRN treats observed expression W as generated from latent TF activities U and a regulatory matrix V = A ⊙ B; the prior over A encodes known or hypothesized TF-gene interactions, and variational inference returns posterior means and variances for every edge. The variances are well calibrated in the sense that lower-uncertainty edges are more likely to reproduce gold-standard interactions. GLM-Prior learns to predict whether a TF and a gene interact from their concatenated sequences using a transformer-based genomic language model; a model trained on one or severa

What carries the argument

The load-bearing object is the prior-knowledge matrix that anchors factorization to named transcription factors. In PMF-GRN, the regulatory matrix is decomposed as V = A ⊙ B, where A is an interaction-existence matrix with a logistic-normal prior whose means come from external evidence; without that prior, the factorization is identifiable only up to column permutation because U V^T is permutation-invariant. Variational inference maximizes a beta-weighted ELBO, and validation AUPRC on held-out prior entries selects hyperparameters. In GLM-Prior, a transformer encoder maps concatenated TF-motif and gene sequences to a pooled embedding, and a binary classifier scores the pair; the scores assem

Load-bearing premise

The load-bearing premise is that TF-gene regulatory interactions can be predicted from nucleotide sequence alone—gene sequence plus TF motif—without chromatin state, cell-type context, or cofactors; if sequence is insufficient for cross-species transfer, the GLM-Prior strategy collapses (the thesis itself acknowledges this in its background chapter).

What would settle it

Train GLM-Prior with the TF-binding motifs replaced by length-matched random sequence while keeping labels and gene sequences fixed, then evaluate on held-out species. If AUPRC does not fall substantially, the classifier is exploiting gene-sequence or label artifacts rather than TF-gene sequence compatibility, and the sequence-prior claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Practitioners can replace heuristic model selection with a single variational objective and a validation AUPRC, so different generative models become comparable on the same footing.
  • Inferred networks carry per-edge uncertainty that is well calibrated: low-variance edges are more accurate, which helps prioritize predictions when gold standards are incomplete.
  • Priors built from sequence alone remove the dependence on cell-type-specific chromatin and ChIP assays, extending GRN inference to less-characterized systems.
  • A model trained on one species or jointly on several transfers to other species, so regulatory knowledge from well-studied organisms can scaffold inference elsewhere.
  • Coupling the sequence prior with probabilistic expression-based inference yields higher accuracy than either alone across yeast, mouse, and human cell lines, supporting the dual-stage design.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If sequence-derived priors truly transfer across species, the same scaffold could be applied to non-model organisms and clinical samples lacking chromatin data; the thesis does not test this directly.
  • The three-by-three cross-comparison of priors and inferers suggests the two stages are modular—improving either stage should improve the whole pipeline—a design principle the thesis demonstrates but does not state as a general claim.
  • Calibrated posterior variances could double as an active-learning signal for choosing which edges to validate experimentally; the thesis stops at calibration and does not explore this use.
  • The synthetic-trajectory experiments show the factorization struggles with long linear cascades, hinting that explicit temporal structure is the natural next extension rather than a refutation of the approach.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The thesis develops two complementary frameworks for gene regulatory network (GRN) inference. Chapter 3 presents PMF-GRN, a probabilistic matrix factorization model with variational inference, which decomposes single-cell expression into transcription factor activity and TF–gene interaction matrices, incorporates prior knowledge through the prior on the interaction matrix, performs hyperparameter selection via a held-out portion of the prior, and outputs posterior uncertainty estimates. Chapter 4 introduces GLM-Prior, a sequence-based prior constructed by fine-tuning a nucleotide transformer to classify TF–gene pairs as interacting or non-interacting, with experiments in yeast, mouse, and human, including single-species training, cross-species transfer, and multi-species training. The thesis argues for a dual-stage view in which sequence-derived priors provide a transferable scaffold and probabilistic inference refines the network with uncertainty quantification. The empirical chapters benchmark against Inferelator, SCENIC, CellOracle, and BEELINE synthetic networks.

Significance. If the claims are borne out, the work makes useful contributions: PMF-GRN decouples generative modeling from inference, provides a principled model-selection procedure, and produces uncertainty estimates that appear well-calibrated in yeast. GLM-Prior addresses a real bottleneck—constructing priors without cell-type-specific assays—and the multi-species experiments are ambitious. The thesis also ships a large amount of experimental work across organisms and synthetic benchmarks, and it is commendably explicit about limitations such as the long-linear synthetic case and the acknowledged incompleteness of gold standards. However, the current evidence does not fully support the strongest claims: the abstract's assertion that PMF-GRN is 'more accurate' than state-of-the-art methods is contradicted by the paper's own figures showing competitive rather than uniformly superior performance, and the cross-species transfer claim for GLM-Prior lacks a homology-aware leakage control for the mouse↔human experiments that are its main evidence.

major comments (3)
  1. [§3.1 Abstract; §3.3.3, Fig. 3.2A; §3.3.6, Fig. 3.5B] The abstract states that 'PMF-GRN infers GRNs more accurately than current state-of-the-art single-cell GRN inference methods,' but the results do not support this uniformly. In Fig. 3.2A, PMF-GRN is competitive with BBSR and CellOracle rather than clearly better; in the 5-fold CV comparison it is similar to BBSR and StARS; and in the BEELINE long-linear synthetic networks (Fig. 3.5B) PMF-GRN is explicitly non-competitive. The wording should be tempered to 'competitive with, and often better than' and the per-dataset exceptions should be stated. This is not merely a presentational issue: the abstract is the central claim of the chapter.
  2. [§3.3.1, Fig. 3.1C; §3.3.3, Fig. 3.2; §3.4.8] The evaluation of PMF-GRN mixes prior-recovery with true discovery in a way that is not quantified. Hyperparameters are selected by AUPRC against a held-out 20% of the prior matrix (Fig. 3.1C), and the final AUPRC is then computed against a 'gold standard.' If the gold standard shares many edges with the prior (as is likely for database- and ChIP-derived references versus motif/accessibility priors), tuning on the prior directly inflates the reported AUPRC because the model is optimized to recover prior-known edges. The manuscript does not report the overlap between the prior and the gold standard, nor performance restricted to gold-standard edges absent from the prior. Please quantify this overlap and report AUPRC on prior-absent edges, or otherwise demonstrate that the reported gains are not primarily prior-recovery.
  3. [§4.3.3, Figs. 4.4–4.5; Supplement Fig. 4.12] The species-transfer claim for GLM-Prior depends on the mouse↔human transfer experiments, but the only homology-aware leakage control (hashFrag) is reported for yeast gene and TF splits (Supplement Fig. 4.12). No equivalent pruning or leakage analysis is described for the mouse→human and human→mouse transfer or multi-species experiments. Because human and mouse share substantial orthologous promoter and TF sequences, a model trained on mouse could score human TF–gene pairs highly merely by recognizing near-identical sequence, without learning a generalizable regulatory code. The transfer AUPRC gains in Figs. 4.4–4.5 could therefore be inflated by sequence homology. This is load-bearing for the dual-stage thesis, which depends on GLM-Prior being genuinely transferable. Please apply hashFrag-based homology pruning to the transfer and multi-species training/test splits, or otherwise quantif
minor comments (4)
  1. [§2.4.4.3, Eq. (2.5)] Equation (2.5) is written as p(W|U,V,σ) = p(U)p(V)p(σ), which equates a conditional likelihood with prior distributions. This should be corrected to define the prior p(U,V,σ) and the likelihood separately, e.g., p(U,V,σ) = p(U)p(V)p(σ).
  2. [Various] The text contains several typos and inconsistent spellings: 'Sacchromyces cerevisiae' (Chapter 3), 'Cell Oracle' versus 'CellOracle', 'releationships' in Fig. 2.4 caption, 'Seqence Language Models' as a section heading, and 'messanger RNA'. These should be corrected in a final pass.
  3. [§4.3.3 and Supplement Fig. 4.12] The hashFrag homology analysis is only in the supplementary material and is not cited in the main text of Section 4.3.3 where transfer learning is described. Add a reference to this control and explicitly note its scope (yeast only) so readers are not misled about the coverage of the leakage analysis.
  4. [§3.4.4 / §3.3.3] The term 'independent gold standard' is used without defining what 'independent' means in relation to the prior. Since the prior and gold standard may share edges, please state explicitly how the gold standard was constructed and what overlap with the prior was measured.

Circularity Check

0 steps flagged

No significant circularity: PMF-GRN and GLM-Prior use held-out evaluation; self-citations are not load-bearing.

full rationale

The thesis's central claims are supported by internally conducted experiments with held-out evaluation. PMF-GRN's hyperparameter selection uses a held-out portion of the prior knowledge matrix as a validation criterion, but the reported performance is evaluated against independent gold standards, and the BEELINE experiment explicitly splits ground truth so that the half used as prior is disjoint from the half used for evaluation. Negative controls with shuffled or absent priors further show that the model is not merely recapitulating its input by construction. GLM-Prior is trained on labeled TF-gene pairs and evaluated on held-out gold standards across cell lines and species; the acknowledged limitation that sequence alone does not encode all regulatory determinants (Section 2.4.4.5) is a scope caveat, not a circular reduction. The self-citations to the author's own prior work ([Skok Gibbs et al. 2022, 2024, 2025]) are cross-references to content reproduced or extended in the thesis, and no load-bearing premise depends solely on an unverified self-citation. The concern that mouse-human transfer experiments lack the hashFrag homology control reported for yeast (Supplement Figure 4.12) is a legitimate evaluation-robustness caveat, but it does not make the reported AUPRC equivalent to the training input by construction. No equation or fitted parameter is reused as its own prediction.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

The central claims rest on the domain assumptions listed above. No new physical or biological entities are postulated; TFA is a latent variable, not a new entity. Free parameters are the hyperparameters tuned to achieve the reported AUPRC.

free parameters (4)
  • PMF-GRN prior strength σ_a = tuned via validation AUPRC
    Controls the strength of the prior over interaction matrix A; selected by hyperparameter search (Section 3.4.2).
  • PMF-GRN TF activity prior σ_u = tuned
    Controls the prior over latent TF activities U; selected by hyperparameter search (Section 3.4.2).
  • β (beta) in modified ELBO = tuned
    Weights the KL term in the modified ELBO (β-VAE style); selected by hyperparameter search (Section 3.4.2).
  • GLM-Prior class weights / negative downsampling rate = selected by one-epoch sweep
    Used to handle class imbalance in TF-target interaction classification; tuned per dataset (Section 4.3.1, Figure 4.2A).
axioms (3)
  • domain assumption TF-gene regulatory interactions are predictable from nucleotide sequence alone (gene sequence + TF motif), without chromatin context or cell-type-specific data.
    GLM-Prior relies on this premise; the thesis itself states in Section 2.4.4.5 that sequence alone misses many regulatory determinants.
  • domain assumption Observed gene expression is a linear combination of latent TF activities, W ≈ U V^T with V = A ⊙ B, under Gaussian noise.
    PMF-GRN's generative model (Section 3.4.1); the linear assumption may be too simple for complex regulation.
  • domain assumption Reference gold standards are sufficiently complete to benchmark GRN inference methods.
    The thesis acknowledges gold standards are incomplete (Section 2.4.2.5); without them, AUPRC comparisons are only partially reliable.

pith-pipeline@v1.3.0-alltime-deepseek · 46264 in / 8886 out tokens · 76796 ms · 2026-08-01T21:28:42.100089+00:00 · methodology

0 comments
read the original abstract

Gene regulatory networks (GRNs) link transcription factor (TF) proteins to their target genes, yet reconstructing these networks from genome-wide data remains challenging under practical and methodological constraints. Many methods couple modeling assumptions to a specific inference procedure and rely on heuristic model selection, while evaluation is constrained by incomplete reference networks and point-estimate outputs that lack uncertainty. GRN reconstruction also depends on prior knowledge to constrain TF-gene interactions, yet available priors are often assay-dependent and difficult to transfer across species and less-characterized systems. In this thesis, we develop two complementary frameworks that address these limitations. In the first, PMF-GRN casts GRN inference as a probabilistic graphical model optimized by variational inference, enabling principled model selection and uncertainty-aware edge estimates. In the second, GLM-Prior addresses the prior bottleneck by fine-tuning the pretrained Nucleotide Transformer to predict TF-target gene interactions directly from nucleotide sequence, while generalizing across yeast, mouse, and human settings. Together, PMF-GRN and GLM-Prior motivate a dual-stage view of GRN reconstruction in which sequence-derived priors provide a transferable starting scaffold and probabilistic inference refines regulatory estimates with quantified uncertainty under incomplete evaluation resources.

Figures

Figures reproduced from arXiv: 2607.16053 by Claudia Skok Gibbs.

Figure 2.1
Figure 2.1. Figure 2.1: The central dogma of biology, illustrating the flow of genetic information from DNA to RNA via transcription and from RNA to protein via translation. Among the many types of proteins produced in the cell, transcription factors (TFs) play a central role in controlling gene regulation. TFs are DNA-binding proteins that recognize specific sequence motifs and modulate transcription by influencing the recruit… view at source ↗
Figure 2.2
Figure 2.2. Figure 2.2: Schematic linking the central dogma to gene regulatory networks. A gene is transcribed into mRNA and translated into protein; when this protein is a transcription factor, it binds to DNA to regulate the transcription of a target gene, initiating a subsequent round of transcription and translation. This repeated feedback process is summarized as a directed graph, where nodes represent genes (or TFs) and d… view at source ↗
Figure 2.3
Figure 2.3. Figure 2.3: Schematic overview of data modalities commonly used to reconstruct gene regulatory net￾works. (A) Microarray workflow: labeled cDNA hybridizes to probes on a fixed array to measure transcript abundance via fluorescence intensity. (B) Bulk RNA-seq: population-level transcriptomes are sequenced to quantify gene expression across conditions or samples. (C) Single-cell RNA-seq: RNA from individual cells is b… view at source ↗
Figure 2.4
Figure 2.4. Figure 2.4: Schematic overview of the input and output of GRN inference algorithms. The primary input is a gene expression matrix, and optional secondary input is a prior knowledge matrix. The selected GRN inference algorithm will produce a directed acyclic graph describing predicted releationships between TFs and their target genes. While a GRN describes direct regulatory interactions between TFs and their target g… view at source ↗
Figure 2.5
Figure 2.5. Figure 2.5: Snapshot sampling and temporal mismatch between TF transcripts and target-gene expres￾sion. Schematic time course illustrating how TF mRNA abundance can be offset in time from the regula￾tory events that drive target gene transcription. At𝑡0, TF mRNA is highly expressed and then declines over subsequent time points, while TF protein accumulates and becomes positioned for regulation. Following an activati… view at source ↗
Figure 2.6
Figure 2.6. Figure 2.6: Key challenges in gene regulatory network inference. (A) A central challenge in GRN re￾construction is determining which transcription factors (TFs) regulate which target genes. (B) Regula￾tion is often combinatorial, with multiple TFs acting cooperatively or competitively to control the same gene, while individual TFs may regulate many targets. (C) Regulatory interactions are also cell-type and context-… view at source ↗
Figure 2.7
Figure 2.7. Figure 2.7: Typical pipeline for constructing prior-knowledge matrix for downstream GRN inference. (A) Open chromatin regions are identified from ATAC-seq data to define candidate regulatory DNA accessible to transcription factor (TF) binding. (B) These accessible regions are scanned for known TF binding motifs to identify potential TF occupancy. (C) Motif-containing accessible regions are assigned to nearby genes, … view at source ↗
Figure 3.1
Figure 3.1. Figure 3.1: (A) PMF-GRN graphical model overview. Input single-cell gene expression 𝑊 is decom￾posed into several latent factors. Information obtained from chromatin accessibility data or genomics databases is incorporated into the prior distribution for 𝐴. (B) Input experimental data for PMF-GRN in￾cludes single-cell RNA-seq gene expression data. Prior-known TF-target gene interactions can be obtained using chromat… view at source ↗
Figure 3.2
Figure 3.2. Figure 3.2: GRN inference in S. cerevisiae. (A) Consensus Network AUPR with a normal prior-knowledge matrix (N): PMF-GRN (red) performance compared to Inferelator algorithms (AMuSR in yellow, BBSR in orange, StARS in green), SCENIC (blue), and CellOracle (purple). Dashed line represents the baseline if expression data is combined. Negative controls: no prior information (NP - black) and shuffled prior information (S… view at source ↗
Figure 3.3
Figure 3.3. Figure 3.3: (A) GRNs inferred by downsampling S. cerevisiae expression data. (B) Hyperparameter search performed on 4 different ratios of cross-validation. Dots represent validation AUPRC from hyperparame￾ter search during cross-validation, triangle represents AUPRC from a GRN learned using the most optimal hyperparameters for each ratio. To underscore the identifiability issue and affirm the utility of prior-known … view at source ↗
Figure 3.4
Figure 3.4. Figure 3.4: GRN and TFA inference in PBMC. (A) UMAP projection of predicted TFA for each annotated PBMC cell type. (B) Predicted IRF2 TFA demonstrates high activity in NK and CD8 T cells. (C) Heat-map dot-plot depicting TFA of selected immune TFs across annotated PBMC cell types. (D) GRN between IRF TFs and their targets. Pink edges indicate literature support for interaction. (E) Heat-map dot-plot indicates ten mos… view at source ↗
Figure 3.5
Figure 3.5. Figure 3.5: PMF-GRN performance on BEELINE synthetic GRN data (A) PMF-GRN inference perfor￾mance with half of the ground truth provided as prior network information and the remaining half pro￾vided as a gold standard for evaluation. Dashed lines are the expected baseline of a random predictor. (B) AUPRC ratio over the baseline random predictor for PMF-GRN in comparison to each of the GRN inference methods used in th… view at source ↗
Figure 3.1
Figure 3.1. Figure 3.1: Results for GRNs learned in B. subtilis datasets B1 (GSE27219) and B2 (GSE67023) compar￾ing "No Normalization" to "Min-Max Scaling". Colored dots represent the normal (N) GRN, with the line indicating the mean of the cross-validation experiments ± standard deviation. Negative controls are demonstrated by black dots for GRNs inferred with No Prior (NP) and grey dots for Shuffled Prior (S). B. subtilis Cro… view at source ↗
Figure 3.2
Figure 3.2. Figure 3.2: PBMC GRN graphs for the family of TFs belonging to (A) SMAD, (B) STAT, (C) GATA, and (D) EGR. TFs for each graph are represented by orange nodes, while target genes are blue nodes. Pink regulatory edges represent interactions with supporting literature. Color scale for blue regulatory edges are scaled from light blue (less out-degree regulation) to darker blue (more out-degree regulation). 101 [PITH_FUL… view at source ↗
Figure 3.3
Figure 3.3. Figure 3.3: UMAP of predicted PBMC TFA. Each UMAP highlights the specific location of activity for each TF considered from the immune PBMC TFs. The final UMAP serves as a reference to TFA annotated by cell-type. 102 [PITH_FULL_IMAGE:figures/full_fig_p126_3_3.png] view at source ↗
Figure 4.1
Figure 4.1. Figure 4.1: Schematic of the dual-stage training pipeline. (A) GLM-Prior is trained on concatenated TF binding motifs and gene sequences labeled as interacting or non-interacting. Input data is split into train￾ing and validation sets, where the training set is downsampled and placed into even class batches using a custom DataLoader. These input sequences are embedded and encoded by the transformer architecture, and… view at source ↗
Figure 4.2
Figure 4.2. Figure 4.2: GLM-Prior model performance across yeast, hESC, and HepG2 cell lines. (A) One-epoch hyperparameter sweep over class-weights and negative-class downsampling rates, evaluated by F1-score. (B) Validation metrics during 10 epoch training using the best hyperparameter configuration. (C) Test AUPRC on held-out gold standards (with chance performance shown for reference in gray). (D) Train and test split compos… view at source ↗
Figure 4.3
Figure 4.3. Figure 4.3: GLM-Prior model performance across mESC, mDC, and mHSC cell lines. (A) One-epoch hyperparameter sweep over class-weights and negative-class downsampling rates, evaluated by F1-score. (B) Validation metrics during 10 epoch training using the best hyperparameter configuration. (C) Test AUPRC on held-out gold standards (with chance performance shown for reference in gray). (D) Train and test split compositi… view at source ↗
Figure 4.4
Figure 4.4. Figure 4.4: Transfer learning and multi-species training with GLM-Prior. (A) Schematic and AUPRC results of the species-transfer learning setup, showing three configurations: (𝑖) a model trained jointly on human and mouse data evaluated on yeast, (𝑖𝑖) a mouse-only model evaluated on human cell lines, and (𝑖𝑖𝑖) a human-only model evaluated on mouse cell lines. (B) Schematic and AUPRC results of the multi-species mode… view at source ↗
Figure 4.5
Figure 4.5. Figure 4.5: Transfer learning and multi-species training with GLM-Prior continued. (A) Comparison of training composition (number of genes and TFs) for single-species, species-transfer, and multi-species models. (B) Class composition (positive and negative labels) in the corresponding training sets. (C) Nor￾malized AUPRC across single-species, species-transfer, and multi-species models for each cell line. 126 [PITH… view at source ↗
Figure 4.6
Figure 4.6. Figure 4.6: Integration of GLM-Prior with PMF-GRN for downstream GRN inference. (A) AUPRC com￾parison of GLM-Prior alone versus integrated with PMF-GRN across six cell line contexts. (B) Edge flip analysis quantifying changes made to the prior knowledge interaction matrix during GRN inference. Blue bars indicate edges gained and red bars indicate edges removed by GRN inference. These findings indicate that when a se… view at source ↗
Figure 4.7
Figure 4.7. Figure 4.7: Comparison of prior construction strategies across species. (A) AUPRC of GLM-Prior, Inferelator-Prior and CellOracle Prior across six cell lines, with chance performance (gray) provided for each context. (B) Normalized AUPRC values with the best-performing prior (green box) highlighted in each cell line. 132 [PITH_FULL_IMAGE:figures/full_fig_p156_4_7.png] view at source ↗
Figure 4.8
Figure 4.8. Figure 4.8: Paired prior-GRN strategies reveal prior dependent gains from expression-based GRN infer￾ence across six cell lines. (A) AUPRC of each prior (GLM-Prior, Inferelator-Prior, and CellOracle’s prior) and its corresponding GRN (PMF-GRN, Inferelator, CellOracle) across six cell lines, with gray dashed lines indicating the chance level in each species. (B) Change in AUPRC from prior to GRN (ΔAUPRC) for each pai… view at source ↗
Figure 4.9
Figure 4.9. Figure 4.9: Paired prior-GRN strategies reveal prior dependent gains from expression-based GRN infer￾ence across six cell lines continued. (A) Normalized AUPRC performance for priors and GRNs in each cell line. Green boxes highlight the best prior and blue boxes highlight the best GRN per cell line. (B) Edge flips between prior and GRN for each method demonstrating how GRN inference methods modulate their priors. Th… view at source ↗
Figure 4.10
Figure 4.10. Figure 4.10: Cross-method comparison disentangles prior and GRN inference contributions. (A) AUPRC for all nine combinations of prior and GRN inference methods across six cell lines. Gray dashed lines indicate chance performance, while light blue, light green, and light pink dashed lines represent baseline performance of GLM-Prior, Inferelator-Prior, and CellOracle prior, respectively, across each cell line. 138 [P… view at source ↗
Figure 4.11
Figure 4.11. Figure 4.11: Cross-method comparison disentangles prior and GRN inference contributions continued. (A) Normalized AUPRC of all nine combinations of prior and GRN for directly comparing across method and cell line. Blue boxes indicate highest GRN performance per cell line. Similar patterns appear in the mammalian cell lines. In hESC, GLM-Prior and Inferelator￾Prior yield the strongest priors (0.19 and 0.17 AUPRC, res… view at source ↗
Figure 4.12
Figure 4.12. Figure 4.12: Sequence-homology analysis of yeast gene and TF splits using hashFrag. (A) [PITH_FULL_IMAGE:figures/full_fig_p180_4_12.png] view at source ↗
Figure 4.13
Figure 4.13. Figure 4.13: Performance Metrics. Additional performance metrics for GLM-Prior and PMF-GRN across six cell lines. 164 [PITH_FULL_IMAGE:figures/full_fig_p188_4_13.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

300 extracted references · 2 canonical work pages

  1. [1]

    Biosystems , volume=

    Gene regulatory network inference: data integration in dynamic models—a review , author=. Biosystems , volume=. 2009 , publisher=

    2009

  2. [2]

    Computers in biology and medicine , volume=

    A review on the computational approaches for gene regulatory network construction , author=. Computers in biology and medicine , volume=. 2014 , publisher=

    2014

  3. [3]

    Bioinformatics , volume=

    Learning gene regulatory networks from gene expression measurements using non-parametric molecular kinetics , author=. Bioinformatics , volume=. 2009 , publisher=

    2009

  4. [4]

    Bioinformatics , volume=

    Inferring quantitative models of regulatory networks from expression data , author=. Bioinformatics , volume=. 2004 , publisher=

    2004

  5. [5]

    arXiv preprint arXiv:1902.08138 , year=

    A nonparametric multi-view model for estimating cell type-specific gene regulatory networks , author=. arXiv preprint arXiv:1902.08138 , year=

    Pith/arXiv arXiv 1902

  6. [6]

    Cell , volume=

    A validated regulatory network for Th17 cell specification , author=. Cell , volume=. 2012 , publisher=

    2012

  7. [7]

    Proceedings of the National Academy of Sciences , volume=

    Inflammatory regulatory network mediated by the joint action of NF-kB, STAT3, and AP-1 factors is involved in many human cancers , author=. Proceedings of the National Academy of Sciences , volume=. 2019 , publisher=

    2019

  8. [8]

    Nature , volume=

    Dynamic regulatory network controlling TH17 cell differentiation , author=. Nature , volume=. 2013 , publisher=

    2013

  9. [9]

    Nature genetics , volume=

    Module networks: identifying regulatory modules and their condition-specific regulators from gene expression data , author=. Nature genetics , volume=. 2003 , publisher=

    2003

  10. [10]

    Biochimica et Biophysica Acta (BBA)-Gene Regulatory Mechanisms , volume=

    Gene regulatory network inference resources: A practical overview , author=. Biochimica et Biophysica Acta (BBA)-Gene Regulatory Mechanisms , volume=. 2020 , publisher=

    2020

  11. [11]

    Genome research , volume=

    Predictive regulatory models in Drosophila melanogaster by integrative inference of transcriptional networks , author=. Genome research , volume=. 2012 , publisher=

    2012

  12. [12]

    Genome biology , volume=

    The Inferelator: an algorithm for learning parsimonious regulatory networks from systems-biology data sets de novo , author=. Genome biology , volume=. 2006 , publisher=

    2006

  13. [13]

    2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society , pages=

    The Inferelator 2.0: a scalable framework for reconstruction of dynamic regulatory network models , author=. 2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society , pages=. 2009 , organization=

    2009

  14. [14]

    BMC bioinformatics , volume=

    Defining transcriptional networks through integrative modeling of mRNA expression and transcription factor binding data , author=. BMC bioinformatics , volume=. 2004 , publisher=

    2004

  15. [15]

    Human heredity , volume=

    Biophysically motivated regulatory network inference: progress and prospects , author=. Human heredity , volume=. 2016 , publisher=

    2016

  16. [16]

    Science , volume=

    Microbial single-cell RNA sequencing by split-pool barcoding , author=. Science , volume=. 2021 , publisher=

    2021

  17. [17]

    2006 , publisher=

    Scalable optimization via probabilistic modeling , author=. 2006 , publisher=

    2006

  18. [18]

    arXiv preprint arXiv:1610.09787 , year=

    Edward: A library for probabilistic modeling, inference, and criticism , author=. arXiv preprint arXiv:1610.09787 , year=

    Pith/arXiv arXiv

  19. [19]

    2015 2nd International Conference on Electronics and Communication Systems (ICECS) , pages=

    Survey on recommendation system methods , author=. 2015 2nd International Conference on Electronics and Communication Systems (ICECS) , pages=. 2015 , organization=

    2015

  20. [20]

    Procedia Computer Science , volume=

    Matrix factorization model in collaborative filtering algorithms: A survey , author=. Procedia Computer Science , volume=. 2015 , publisher=

    2015

  21. [21]

    Proceedings of the 10th international conference on World Wide Web , pages=

    Item-based collaborative filtering recommendation algorithms , author=. Proceedings of the 10th international conference on World Wide Web , pages=

  22. [22]

    Computer , volume=

    Matrix factorization techniques for recommender systems , author=. Computer , volume=. 2009 , publisher=

    2009

  23. [23]

    data Science (Second Edition) , pages=

    Recommendation engines , author=. data Science (Second Edition) , pages=

  24. [24]

    Heredity , volume=

    Epigenetics and gene expression , author=. Heredity , volume=. 2010 , publisher=

    2010

  25. [25]

    Annual review of cell and developmental biology , volume=

    Comparative analysis of gene regulatory networks: from network reconstruction to evolution , author=. Annual review of cell and developmental biology , volume=. 2015 , publisher=

    2015

  26. [26]

    Current opinion in biotechnology , volume=

    Network-based approaches for analysis of complex biological systems , author=. Current opinion in biotechnology , volume=. 2016 , publisher=

    2016

  27. [27]

    Cell , volume=

    Comprehensive integration of single-cell data , author=. Cell , volume=. 2019 , publisher=

    2019

  28. [28]

    Bioinformatics , volume=

    Estimating the activity of transcription factors by the effect on their target genes , author=. Bioinformatics , volume=. 2014 , publisher=

    2014

  29. [29]

    The Journal of Machine Learning Research , volume=

    Pyro: Deep universal probabilistic programming , author=. The Journal of Machine Learning Research , volume=. 2019 , publisher=

    2019

  30. [30]

    arXiv preprint arXiv:1912.11554 , year=

    Composable effects for flexible and accelerated probabilistic programming in NumPyro , author=. arXiv preprint arXiv:1912.11554 , year=

    Pith/arXiv arXiv 1912

  31. [31]

    2012 IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB) , pages=

    Matrix factorization for transcriptional regulatory network inference , author=. 2012 IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB) , pages=. 2012 , organization=

    2012

  32. [32]

    elife , volume=

    A new protocol for single-cell RNA-seq reveals stochastic gene expression during lag phase in budding yeast , author=. elife , volume=. 2020 , publisher=

    2020

  33. [33]

    Nature , volume=

    Neuronal diversity and convergence in a visual system developmental atlas , author=. Nature , volume=. 2021 , publisher=

    2021

  34. [34]

    Neuron , volume=

    Transcriptional programs of circuit assembly in the Drosophila visual system , author=. Neuron , volume=. 2020 , publisher=

    2020

  35. [35]

    Cell reports , volume=

    Condition-specific modeling of biophysical parameters advances inference of regulatory networks , author=. Cell reports , volume=. 2018 , publisher=

    2018

  36. [36]

    Nucleic acids research , volume=

    YEASTRACT+: a portal for cross-species comparative genomics of transcription regulation in yeasts , author=. Nucleic acids research , volume=. 2020 , publisher=

    2020

  37. [37]

    Nucleic acids research , volume=

    YEASTRACT: an upgraded database for the analysis of transcription regulatory networks in Saccharomyces cerevisiae , author=. Nucleic acids research , volume=. 2018 , publisher=

    2018

  38. [38]

    Frontiers in Microbiology , volume=

    Reconstruction of the regulatory network for Bacillus subtilis and reconciliation with gene expression data , author=. Frontiers in Microbiology , volume=. 2016 , publisher=

    2016

  39. [39]

    Nature , pages=

    Decoding gene regulation in the fly brain , author=. Nature , pages=. 2022 , publisher=

    2022

  40. [40]

    Nature communications , volume=

    Characterizing chromatin landscape from aggregate and single-cell genomic assays using flexible duration modeling , author=. Nature communications , volume=. 2020 , publisher=

    2020

  41. [41]

    Nucleic acids research , volume=

    MEME SUITE: tools for motif discovery and searching , author=. Nucleic acids research , volume=. 2009 , publisher=

    2009

  42. [42]

    Artificial intelligence and statistics , pages=

    Black box variational inference , author=. Artificial intelligence and statistics , pages=. 2014 , organization=

    2014

  43. [43]

    Irina Higgins and Loic Matthey and Arka Pal and Christopher Burgess and Xavier Glorot and Matthew Botvinick and Shakir Mohamed and Alexander Lerchner , booktitle=. beta-. 2017 , url=

    2017

  44. [44]

    Proceedings of the National Academy of Sciences , volume=

    Singular value decomposition for genome-wide expression data processing and modeling , author=. Proceedings of the National Academy of Sciences , volume=. 2000 , publisher=

    2000

  45. [45]

    Bioinformatics , volume=

    Application of Bayesian decomposition for analysing microarray data , author=. Bioinformatics , volume=. 2002 , publisher=

    2002

  46. [46]

    Genome research , volume=

    Subsystem identification through dimensionality reduction of large-scale gene expression data , author=. Genome research , volume=. 2003 , publisher=

    2003

  47. [47]

    Proceedings of the national academy of sciences , volume=

    Metagenes and molecular pattern discovery using matrix factorization , author=. Proceedings of the national academy of sciences , volume=. 2004 , publisher=

    2004

  48. [48]

    Bioinformatics , volume=

    Improving molecular cancer class discovery through sparse non-negative matrix factorization , author=. Bioinformatics , volume=. 2005 , publisher=

    2005

  49. [49]

    Bioinformatics , volume=

    A non-negative matrix factorization method for detecting modules in heterogeneous omics multi-modal data , author=. Bioinformatics , volume=. 2016 , publisher=

    2016

  50. [50]

    Proceedings of the National Academy of Sciences , volume=

    Integrative analysis of single-cell genomics data by coupled nonnegative matrix factorizations , author=. Proceedings of the National Academy of Sciences , volume=. 2018 , publisher=

    2018

  51. [51]

    PLoS computational biology , volume=

    Building gene regulatory networks from scATAC-seq and scRNA-seq using linked self organizing maps , author=. PLoS computational biology , volume=. 2019 , publisher=

    2019

  52. [52]

    Current Opinion in Systems Biology , volume=

    Gene regulatory network inference in single-cell biology , author=. Current Opinion in Systems Biology , volume=. 2021 , publisher=

    2021

  53. [53]

    Genome biology , volume=

    Eleven grand challenges in single-cell data science , author=. Genome biology , volume=. 2020 , publisher=

    2020

  54. [54]

    Science , volume=

    Condition-dependent transcriptome reveals high-level regulatory architecture in Bacillus subtilis , author=. Science , volume=. 2012 , publisher=

    2012

  55. [55]

    Nucleic acids research , volume=

    Subti Wiki 2.0—an integrated database for the model organism Bacillus subtilis , author=. Nucleic acids research , volume=. 2016 , publisher=

    2016

  56. [56]

    Nucleic acids research , volume=

    Subti Wiki in 2018: from genes and proteins to functional network annotation of the model organism Bacillus subtilis , author=. Nucleic acids research , volume=. 2018 , publisher=

    2018

  57. [57]

    Nucleic Acids Research , volume=

    The current state of Subti Wiki, the database for the model organism Bacillus subtilis , author=. Nucleic Acids Research , volume=. 2022 , publisher=

    2022

  58. [58]

    Bioinformatics , volume=

    Inferring gene regulatory networks from multiple microarray datasets , author=. Bioinformatics , volume=. 2006 , publisher=

    2006

  59. [59]

    Bioinformatics , volume=

    Fast network component analysis (FastNCA) for gene regulatory network reconstruction from microarray data , author=. Bioinformatics , volume=. 2008 , publisher=

    2008

  60. [60]

    Cell , volume=

    Integrated analysis of multimodal single-cell data , author=. Cell , volume=. 2021 , publisher=

    2021

  61. [61]

    bioRxiv , pages=

    Structure primed embedding on the transcription factor manifold enables transparent model architectures for gene regulatory network and latent activity inference , author=. bioRxiv , pages=. 2023 , publisher=

    2023

  62. [62]

    Frontiers in immunology , volume=

    IRF2 is required for development and functional maturation of human NK cells , author=. Frontiers in immunology , volume=. 2022 , publisher=

    2022

  63. [63]

    Immunity , volume=

    The transcription factor IRF2 drives interferon-mediated CD8+ T cell exhaustion to restrict anti-tumor immunity , author=. Immunity , volume=. 2022 , publisher=

    2022

  64. [64]

    Blood, The Journal of the American Society of Hematology , volume=

    Regulation of human 2-microglobulin transactivation in hematopoietic cells , author=. Blood, The Journal of the American Society of Hematology , volume=. 2003 , publisher=

    2003

  65. [65]

    PLoS One , volume=

    Immunopathology of childhood celiac disease—Key role of intestinal epithelial cells , author=. PLoS One , volume=. 2017 , publisher=

    2017

  66. [66]

    Nature communications , volume=

    IRF2 is a master regulator of human keratinocyte stem cell fate , author=. Nature communications , volume=. 2019 , publisher=

    2019

  67. [67]

    Cell Reports , volume=

    A Golgi-resident GPR108 cooperates with E3 ubiquitin ligase Smurf1 to suppress antiviral innate immunity , author=. Cell Reports , volume=. 2023 , publisher=

    2023

  68. [68]

    Immunity , volume=

    The ubiquitin ligase RNF5 regulates antiviral responses by mediating degradation of the adaptor protein MITA , author=. Immunity , volume=. 2009 , publisher=

    2009

  69. [69]

    Biomolecules & Therapeutics , volume=

    Identification of small GTPases that phosphorylate IRF3 through TBK1 activation using an active mutant library screen , author=. Biomolecules & Therapeutics , volume=. 2023 , publisher=

    2023

  70. [70]

    Nature immunology , volume=

    Pax5: the guardian of B cell identity and function , author=. Nature immunology , volume=. 2007 , publisher=

    2007

  71. [71]

    Nature immunology , volume=

    The contribution of transcription factor IRF1 to the interferon- --interleukin 12 signaling axis and TH1 versus TH-17 differentiation of CD4+ T cells , author=. Nature immunology , volume=. 2008 , publisher=

    2008

  72. [72]

    Immunology , volume=

    SMAD regulatory networks construct a balanced immune system , author=. Immunology , volume=. 2013 , publisher=

    2013

  73. [73]

    Nature methods , volume=

    Benchmarking algorithms for gene regulatory network inference from single-cell transcriptomic data , author=. Nature methods , volume=. 2020 , publisher=

    2020

  74. [74]

    Genome biology , volume=

    SCANPY: large-scale single-cell gene expression data analysis , author=. Genome biology , volume=. 2018 , publisher=

    2018

  75. [75]

    Proceedings of the international AAAI conference on web and social media , volume=

    Gephi: an open source software for exploring and manipulating networks , author=. Proceedings of the international AAAI conference on web and social media , volume=

  76. [76]

    BMC Molecular Biology , volume=

    Properties of STAT1 and IRF1 enhancers and the influence of SNPs , author=. BMC Molecular Biology , volume=. 2017 , publisher=

    2017

  77. [77]

    Jak-stat , volume=

    Transcriptional regulation by STAT1 and STAT2 in the interferon JAK-STAT pathway , author=. Jak-stat , volume=. 2013 , publisher=

    2013

  78. [78]

    Addiction biology , volume=

    Dysregulation of the histone demethylase KDM6B in alcohol dependence is associated with epigenetic regulation of inflammatory signaling pathways , author=. Addiction biology , volume=. 2021 , publisher=

    2021

  79. [79]

    Blood , volume=

    Activation of the STAT1-BCL-2/MCL-1 Axis in Leukemic Cells Carrying a SPAG9-JAK2 Fusion , author=. Blood , volume=. 2021 , publisher=

    2021

  80. [80]

    Immunity , volume=

    Tumor necrosis factor receptor signaling in keratinocytes triggers interleukin-24-dependent psoriasis-like skin inflammation in mice , author=. Immunity , volume=. 2013 , publisher=

    2013

Showing first 80 references.