Pith. sign in

REVIEW 4 major objections 5 minor 62 references

A dual-graph model with domain-aware edge typing and evidential uncertainty outperforms thirteen baselines in predicting spatial gene expression from H&E histology, and its confidence intervals achieve near-exact 90% coverage.

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 ↗

HierarchicalDAEW predicts spatial gene expression with expression-derived domain edge typing and calibrated uncertainty, beating 13 baselines on breast Visium sections.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection Expression-derived edge typing at test time undercuts the H&E-only claim; the paper is thorough but the central result is not supported. the 4 major comments →

arxiv 2607.20896 v1 pith:VKLSPHM4 submitted 2026-07-23 cs.LG q-bio.GN

HierarchicalDAEW: Domain-Aware Edge-Weighted Graph Convolution with Evidential Uncertainty for Multi-Section Spatial Gene Expression Prediction from H&E Histology

Kritanu Chattopadhyay , Soumya Chatterjee , Ondrej Krejcar , Debotosh Bhattacharjee This is my paper
classification cs.LG q-bio.GN
keywords spatial transcriptomicsgene expression predictionH&E histologygraph convolutional networkstissue domainsedge typingevidential uncertaintyVisium
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.

The reading

The paper tries to establish that gene expression at tissue spots can be predicted from H&E histology better when the graph message passing respects tissue domain structure instead of treating all spatial edges alike, and that evidential deep learning gives trustworthy per-prediction confidence intervals. A sympathetic reader would care because if true, this gives pathologists a cheap, assay-free route to spatially resolved molecular profiling, with calibrated flags for when predictions should not be acted on. The central claim is the architecture's performance: it beats thirteen published methods across six human Visium sections and its 90% credible intervals contain the true expression 90.3% of the time, versus 32.3% for Monte Carlo dropout. The paper further claims the benefit is specifically due to expression-derived domain typing, not any typed edges, and reports that a morphology-derived typing is harmful.

Core claim

On its own terms, the discovery is that a graph convolutional operator which types each spatial edge as intra-domain, inter-domain, or boundary—with separate learned projections per type—plus a hierarchical domain-level pooling with gated fusion, produces state-of-the-art correlation between predicted and measured expression (mean PCC 0.696 under multi-section joint training; 0.704 single-section), and that a Normal-Inverse-Gamma evidential head yields confidence intervals with near-exact 90% empirical coverage that remain informative under conformal calibration. The authors also show the gene graph decoder propagates landmark-gene predictions to unsupervized genes (imputation PCC 0.831), an

What carries the argument

The load-bearing mechanism is DAEWConv, a relational-style convolution in which edges in the spot graph are typed by Leiden-derived tissue domains (intra-domain, inter-domain, boundary) and each type has its own learned projection matrix plus a learnable scalar gate; a DomainGCN over pooled domain representations and a per-spot CrossScaleGate fuse coarse tissue context back into spot features. A second gene-level graph with protein–protein interaction priors and co-expression edges propagates landmark predictions via source-typed attention-gated message passing, and a Normal-Inverse-Gamma evidential loss outputs both an expression estimate and a calibrated variance.

Load-bearing premise

The method needs expression-derived tissue domains to type edges, but for a genuinely new H&E slide no expression is available at inference time; the authors assign test domains by nearest-centroid matching in expression-PC space (Eq. 4), and their own ablation shows replacing this with morphology-based typing drops PCC below the untyped baseline (0.598 vs 0.651).

What would settle it

Re-run the edge-typing experiment on a fresh slide with no expression data, using only a morphology-based domain predictor, and check whether mean PCC falls below the untyped GCN control; if it does not, the paper's central premise fails. Also, an independent reproduction of the 0.696 mean PCC on three breast sections, or of the 0.903 vs 0.323 coverage gap versus MC dropout, would directly test the headline numbers.

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

If this is right

  • If correct, this method becomes the strongest published approach for H&E-to-expression prediction under the reported evaluation protocol, with a meaningful edge when training spans multiple sections.
  • The near-exact 90% coverage means evidential NIG can replace Monte Carlo dropout for trustworthy per-spot confidence intervals without extra forward passes.
  • The gene graph decoder extends prediction to genes never directly supervised (imputation PCC 0.831), suggesting scalable expansion of gene panels.
  • The hierarchy ablation (+0.044 PCC) implies domain-level context is a genuine source of predictive signal worth incorporating in future models.
  • The benefit concentrates in multi-section training and depends on expression-derived domains; morphology-derived typing actively hurts.

Where Pith is reading between the lines

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

  • Editorial inference: because test-time domain labels are assigned by nearest-centroid matching in expression-PC space, the method cannot yet be applied directly to an unmeasured slide; the paper's own ablation suggests a morphology-based domain predictor would not recover the gain.
  • Editorial inference: the near-tie against the strongest baseline on single-section breast (PCC 0.704 vs 0.699, not significant) means the claimed advantage is partly contingent on multi-section joint training; single-section gains may not generalize to all tissues.
  • Editorial inference: the calibrated uncertainty degrades sharply under tissue-type shift (coverage 0.333–0.499), so the 'trustworthy intervals' claim should be read as in-distribution only; conformal recalibration on the target tissue would be a cheap practical fix.
  • Editorial inference: the success of expression-derived typing suggests that learning tissue-domain structure directly from H&E in a supervised or self-supervised way could close the deployment gap, offering a concrete next step the paper leaves implicit.
Share X Bluesky LinkedIn Reddit HN

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

4 major / 5 minor

Summary. The paper proposes HierarchicalDAEW, a dual-graph architecture for predicting spatial gene expression from H&E histology. The spot-level encoder, DAEWConv, types spatial edges as intra-domain, inter-domain, or boundary based on Leiden domains computed from expression data, and learns separate projections per edge type. A domain-level GCN with cross-scale gating adds hierarchical context, and a gene-graph decoder propagates landmark-gene predictions through STRING-DB and co-expression edges. Uncertainty is modeled with a Normal-Inverse-Gamma evidential loss. The authors report state-of-the-art Pearson correlation against thirteen baselines across six Visium sections, with ablations, negative controls, multi-seed checks, and calibration studies.

Significance. If the central claim were valid, HierarchicalDAEW would be a substantial advance toward H&E-only spatial expression prediction with calibrated uncertainty. The manuscript is unusually thorough in its evaluation: multi-seed reproducibility, bootstrap confidence intervals, per-gene win rates, an untouched holdout, and conformal calibration are all included. However, the core edge-typing mechanism requires expression data at test time through Eq. (4), which is not available for a new slide; the paper's own ablation (Sec. 6.1.5) shows that the morphology-derived proxy is actively harmful. This undermines the stated practical significance and the validity of the reported benchmark gains.

major comments (4)
  1. [Sec. 3.4, Eq. (4)] Test-spot domain labels are assigned by nearest-centroid matching in expression-PC space using the test spot's expression vector z_i. These labels determine edge types in Eq. (5), which are consumed by DAEWConv in Eqs. (6)-(7). Thus, at evaluation time, the model's edge-typing input contains information from the very expression values being predicted. This contradicts the problem formulation in Eq. (2) and the claimed H&E-only input. The negative control in Sec. 6.2.4 shuffles labels but keeps the real expression-derived edge types, so it does not test this form of leakage. A valid control would recompute or permute edge types from shuffled expression.
  2. [Sec. 6.1.5, Table 12] The paper shows that expression-derived typing (PCC 0.683) outperforms no typing (0.651), while morphology-derived typing performs worse than no typing (0.598). Since expression is unavailable for an unseen slide, the deployment-time variant of the model would have to use morphology-derived typing, which would be worse than a standard GCN. The reported multi-section advantage (Table 3: 0.696 vs 0.638) therefore does not carry over to the clinical setting claimed in the abstract and introduction. This is acknowledged in Sec. 6.1.5 as 'an important limitation', but the conclusion and abstract still frame the method as an H&E-only predictor.
  3. [Sec. 4.3, baselines] All baselines are trained with a single learning rate (5e-4) with no per-model tuning, while HierarchicalDAEW's hyperparameters (depth, hidden dimension, neighborhood size, Leiden resolution, loss weights, temperature) are selected via grid search in Sec. 6.1. This asymmetry can inflate the reported margins. At minimum, a learning-rate sensitivity analysis for the strongest baselines (SEPAL, THItoGene, MERGE) is needed to support the claim that HierarchicalDAEW is the strongest method, especially given the non-significant single-section gap to SEPAL.
  4. [Sec. 5.2, Table 4] In the single-section benchmark, HierarchicalDAEW's PCC (0.704) is not statistically distinguishable from SEPAL (0.699). The paper's headline claim of outperforming thirteen published baselines is therefore supported only in the multi-section setting (Table 3), where the expression-derived edge-typing leakage is present. The per-gene win rate in Sec. 6.2.1 does not resolve this aggregate non-significance for the closest competitor; a method that is not significantly better than a baseline on the standard single-section protocol cannot be described as unequivocally strongest.
minor comments (5)
  1. [Sec. 3.4] The heading 'Leakage-Free Leiden Domain Assignment' overstates what the procedure achieves. The nearest-centroid matching avoids training-set contamination, but it still injects test-side expression information into the model inputs via the edge typing. Please reword to reflect that the procedure only prevents fold-level contamination, not target-derived inference leakage.
  2. [Sec. 4.3, TRIPLEX] TRIPLEX is approximated with single-scale UNI embeddings because multi-resolution inputs are unavailable. This should be stated as a limitation in the main experimental section, not only in the implementation details, since it may materially change the baseline's performance.
  3. [Tables 3 and 4] The metric sets differ between Table 3 and Table 4: the former reports PCC, Spearman, R2; the latter adds CCC and drops some entries. Please use a consistent set of metrics across tables to make comparisons easier.
  4. [Sec. 6.4.2] The Spearman/Pearson correlations in the calibration summary (Table 19: 0.176/0.119) differ from those in Figure 30 (0.249/0.249). The text explains that these come from separately trained instances, but the discrepancy should be made explicit in the table or figure caption to avoid confusion.
  5. [Data and code availability] Code is listed as 'available upon acceptance'. For a benchmark paper with strong empirical claims, a public repository at submission time would strengthen reproducibility and is customary for this venue.

Circularity Check

2 steps flagged

Test-time edge typing uses expression (Eq. 4), so reported H&E-only SOTA depends on target-derived inputs; paper's own ablation shows morphology typing is worse than no typing.

specific steps
  1. self definitional [Section 3.4, Eq. (4) with edge typing in Eq. (5)]
    "Domain labels for held-out test spots are then assigned by nearest-centroid matching in principal component space, c_i= argmin_{c∈{1,…,C}} ||z_i−μ_c||_2, i∉train, where z_i is the principal-component representation of test spot i and μ_c is the centroid of training spots assigned to domain c, ensuring that no test-spot expression contributes to domain formation while still providing every spot with a domain label."

    The z_i used in Eq. (4) is the expression-PC embedding of the held-out spot, so the domain label c_i—and therefore the DAEWConv edge type r_ij in Eq. (5)—is a function of the very expression values the model is supposed to predict from H&E. The 'leakage-free' claim only says test expression does not form the Leiden centroids; it still determines the nearest-centroid assignment and hence the model's structural input. Thus the evaluation is not a pure H&E-to-expression prediction; target information enters through the graph edge typing.

  2. fitted input called prediction [Section 6.1.5, Table 12]
    "Expression-derived typing achieves the highest mean PCC (0.683), substantially outperforming both alternatives. Morphology-derived typing performs worst of the three (0.598), falling 0.053 PCC below even the untyped control (0.651), while the untyped control itself trails expression-derived typing by 0.032 PCC. ... Naively substituting a histology-derived proxy for domain structure is actively harmful relative to no typing at all."

    This ablation is the paper's own demonstration that DAEWConv's advantage (0.683 vs 0.651 over no typing, and the margin over all thirteen baselines) comes specifically from expression-derived domain labels, not from morphology that is available on an unseen H&E slide. Since morphology-derived typing is worse than no typing, the reported SOTA gain is attributable to the target-expression-derived edge input of Eq. (4), not to an H&E-only prediction mechanism. The 'expression-derived typing' is a fitted, target-dependent input being credited as an architectural prediction gain.

full rationale

The central claimed result—predicting spatial gene expression from H&E histology at SOTA level—is undermined by a target-derived input. Section 3.4 assigns held-out test spots to Leiden domains by nearest-centroid matching in expression-PC space (Eq. 4), and these labels define the DAEWConv edge types (Eq. 5). At test time the model therefore consumes information computed from the expression values it is supposed to predict. The paper's own edge-typing-source ablation (Sec. 6.1.5) confirms the load-bearing nature of this leakage: expression-derived typing gives PCC 0.683, no typing 0.651, and morphology-derived typing only 0.598. So the advantage of DAEWConv, and hence the margin over the thirteen baselines, is not available when only H&E is used for a new slide. The shuffled-label negative control (Sec. 6.2.4) does not address this: edge types are kept at their real-expression values while only labels are shuffled, so it rules out positional shortcuts but not target-derived inputs. Other parts of the paper (gene-graph imputation, NIG uncertainty calibration) are not circular in the same way and could stand independently, but the headline H&E-only SOTA claim reduces to a setting where the target expression is already partially supplied as structural input. Score is 7 rather than higher because the model still learns a nontrivial histology-to-expression mapping and the other contributions are not equivalence-by-construction; it is not a fully vacuous derivation.

Axiom & Free-Parameter Ledger

8 free parameters · 6 axioms · 0 invented entities

The central claim rests on several fitted hyperparameters selected by ablation on breast tissue, plus domain assumptions about expression-derived domains and gene-graph priors. The most consequential is the nearest-centroid domain assignment for test spots, which requires test expression and is therefore not usable for genuinely new H&E slides.

free parameters (8)
  • Leiden resolution = 0.7
    Selected via ablation over {0.3,0.5,0.7,1.0,1.5} on breast tissue; not derived from theory.
  • Contrastive temperature tau = 0.1
    Selected via ablation over {0.01,0.05,0.07,0.1,0.3}.
  • lambda_NIG = 0.01
    Chosen via hyperparameter search; controls balance between MSE and evidential loss.
  • lambda_ctr = 0.1
    Chosen via hyperparameter search; controls contrastive loss weight.
  • Hidden dimension = 1024
    Selected via ablation; PCC increased monotonically with hidden dimension up to 1024.
  • Neighborhood size K = 4
    Selected via ablation and joint grid search over {4,6,8,10,14,18,24}.
  • Number of DAEWConv layers L = 2
    Selected via ablation; performance peaks at L=2 and degrades with depth.
  • STRING-DB confidence threshold = 900
    Selected via ablation over {500,700,900}; strictest threshold retained.
axioms (6)
  • domain assumption k-NN graph in physical coordinate space adequately captures spatial expression structure.
    Used throughout the spot graph construction (Sec. 3.1); if spatial signal is not local in physical space, the graph is misspecified.
  • domain assumption Leiden clustering on expression PCs yields biologically meaningful tissue domains.
    Edge typing depends on these domains (Sec. 3.4); the paper validates indirectly but does not independently prove domain validity.
  • ad hoc to paper Nearest-centroid assignment of test spots in expression-PC space gives valid, leakage-free domain labels.
    Eq. (4) uses test expression to assign domains; the paper calls this leakage-free but it requires expression at test time, which is the core weakness.
  • domain assumption STRING-DB interactions and co-expression edges propagate expression signal across genes.
    The gene graph decoder relies on these edges (Sec. 3.7); no independent validation that the resulting imputation is biologically causal.
  • domain assumption Normal-Inverse-Gamma evidential loss produces calibrated uncertainty estimates.
    The NIG model assumes a particular higher-order distribution (Sec. 3.8); coverage is measured but the assumption is not tested independently.
  • domain assumption UNI histology embeddings are sufficient features for predicting expression.
    The model uses frozen UNI embeddings (Sec. 3.2); if morphology features lack expression-related signal, the whole pipeline is limited.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of HierarchicalDAEW: Domain-Aware Edge-Weighted Graph Convolution with Evidential Uncertainty for Multi-Section Spatial Gene Expression Prediction from H&E Histology." pith.science (2026). https://pith.science/paper/VKLSPHM4

@misc{pith2026260720896,
  author       = {Pith},
  title        = {Pith review of: HierarchicalDAEW: Domain-Aware Edge-Weighted Graph Convolution with Evidential Uncertainty for Multi-Section Spatial Gene Expression Prediction from H&E Histology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VKLSPHM4}},
  note         = {Machine review of arXiv:2607.20896}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Spatial transcriptomics assays remain costly and technically demanding, restricting transcriptome-wide profiling to specialist settings and preventing routine clinical deployment. Predicting spatially resolved gene expression from H&E histology could close this gap, yet current methods largely ignore the underlying tissue architecture and rarely quantify how their predictions can be trusted. We introduce HierarchicalDAEW, a dual-graph architecture that addresses both gaps. On the spot graph, a Domain-Aware Edge-Weighted convolutional operator learns separate projections for inter-domain, intra-domain, and boundary edges derived from Leiden clustering, allowing the model to treat tissue heterogeneity as an explicit structural signal rather than an implicit one. A second gene-level graph then fuses protein-protein interaction priors from STRING-DB with tissue-specific co-expression through learned attention gating, propagating predictions from a landmark gene set to a broader gene panel. Reliability is handled through evidential uncertainty estimation, which produces far better calibrated confidence intervals than Monte Carlo dropout under identical conditions. Across six human Visium sections spanning breast, colorectal, prostate, and cerebellar tissue, and against thirteen published baselines, HierarchicalDAEW achieves the strongest correlation with ground-truth expression, with gains that hold up under multi-seed reproducibility checks and negative controls that rule out positional shortcuts. Ablations further confirm that both the domain-aware edge typing and the hierarchical depth are necessary to this improvement, and calibrated uncertainty estimates identify low-confidence predictions for pathologist review before clinical action.

Figures

Figures reproduced from arXiv: 2607.20896 by Debotosh Bhattacharjee, Kritanu Chattopadhyay, Ondrej Krejcar, Soumya Chatterjee.

Figure 1
Figure 1. Figure 1: Overview of HierarchicalDAEW. A frozen UNI encoder extracts spot features, which are processed by a domain-aware spot-level encoder (DAEWConv, DomainGCN, CrossScaleGate) to produce Stage-1 landmark predictions, propagated by a gene graph decoder to a broader panel with calibrated evidential uncertainty. The encoder is used strictly as a frozen feature extrac￾tor. Its weights are not updated during training… view at source ↗
Figure 2
Figure 2. Figure 2: DAEWConv edge typing and per-type message [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Gene graph decoder construction. STRING-DB [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Multi-section joint training results on Breast S1 [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Single-section benchmark on Breast S1 (5-fold [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Extended single-section metrics on Breast S1: [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Cross-tissue generalization of HierarchicalDAEW, [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Leave-one-dataset-out (LODO) evaluation: mean [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Bidirectional few-shot adaptation between Breast [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Histology encoder ablation. UNI Chen et al. (2024b) outperforms both ResNet-50 He et al. (2016) and Prov-GigaPath Xu et al. (2024), confirming that histopathology-specific pretraining matters more than en￾coder scale [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: GNN depth ablation. Performance peaks at [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Spatial neighborhood size ablation. Accuracy [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: Architectural and training ablations: fixed vs. [PITH_FULL_IMAGE:figures/full_fig_p016_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Leiden clustering resolution ablation Hu et al. (2021). Performance follows a clear unimodal trend, peaking at resolution 0.7 [PITH_FULL_IMAGE:figures/full_fig_p016_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Contrastive loss temperature ablation Xie et al. (2023). Accuracy is largely insensitive to 𝜏 within the tested range, with 𝜏=0.1 marginally best. Chattopadhyay et al.: Preprint submitted to Elsevier Page 15 of 30 [PITH_FULL_IMAGE:figures/full_fig_p016_16.png] view at source ↗
Figure 18
Figure 18. Figure 18: Joint 𝐿 × 𝐾 grid search results. The greedily se￾lected configuration (𝐿=2, 𝐾=4) is confirmed as the jointly optimal combination [PITH_FULL_IMAGE:figures/full_fig_p017_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Stability of Leiden domain counts across cross [PITH_FULL_IMAGE:figures/full_fig_p017_19.png] view at source ↗
Figure 21
Figure 21. Figure 21: Per-gene win rate of HierarchicalDAEW against [PITH_FULL_IMAGE:figures/full_fig_p018_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Bootstrap Tibshirani and Efron (1993) 95% con￾fidence intervals for mean PCC difference against each base￾line. Only the SEPAL Mejia et al. (2023) comparison crosses zero, consistent with the non-significant result reported in [PITH_FULL_IMAGE:figures/full_fig_p019_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Multi-seed reproducibility across three random [PITH_FULL_IMAGE:figures/full_fig_p020_23.png] view at source ↗
Figure 25
Figure 25. Figure 25: Frobenius distance between DAEWConv’s type [PITH_FULL_IMAGE:figures/full_fig_p021_25.png] view at source ↗
Figure 29
Figure 29. Figure 29: NIG Amini et al. (2020) uncertainty calibration summary. Empirical coverage at the 90% target is 0.903, closely matching the nominal level. 6.4.2. Reliability Diagram and Risk-Coverage Curve We assess calibration quality visually through two com￾plementary diagnostics. The reliability diagram bins held￾out spots by predicted uncertainty 𝜎𝑖 into deciles and plots mean predicted uncertainty against mean abs… view at source ↗
Figure 28
Figure 28. Figure 28: UMAP McInnes et al. (2018) projection of learned spot embeddings on held-out test spots, colored by Leiden domain (left) and by per-spot PCC Benesty et al. (2009) (right). 6.4. Uncertainty Quantification 6.4.1. NIG Calibration: Coverage and ENCE Beyond point-prediction accuracy, we care about whether the model’s uncertainty estimates are actually useful, whether high predicted uncertainty really does corr… view at source ↗
Figure 30
Figure 30. Figure 30: Reliability diagram (left), risk-coverage curve [PITH_FULL_IMAGE:figures/full_fig_p023_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: NIG evidential uncertainty Amini et al. (2020) versus MC Dropout Gal and Ghahramani (2016). NIG achieves near-exact coverage (0.903) and substantially stronger error correlation than MC Dropout, which under￾covers severely (0.323). 6.4.5. Conformal Prediction Guarantees Unlike the NIG-based intervals evaluated so far, which rely on the correctness of the assumed Normal-Inverse￾Gamma predictive distributio… view at source ↗
Figure 32
Figure 32. Figure 32: Split conformal prediction Stephen et al. (2021) coverage (left) and average interval width (right) across target levels. Empirical coverage tracks the target closely at every level tested. 6.4.6. Calibration Under Dataset Shift The calibration results reported so far are evaluated in￾distribution, on held-out spots from the same section, or a section of the same tissue type, used during model training an… view at source ↗
Figure 34
Figure 34. Figure 34: Empirical scalability with spot count: inference [PITH_FULL_IMAGE:figures/full_fig_p026_34.png] view at source ↗
Figure 35
Figure 35. Figure 35: Top significantly enriched KEGG and GO Bi [PITH_FULL_IMAGE:figures/full_fig_p027_35.png] view at source ↗
Figure 36
Figure 36. Figure 36: Final untouched holdout evaluation: per-spot [PITH_FULL_IMAGE:figures/full_fig_p028_36.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

62 extracted references · 8 linked inside Pith

  1. [1]

    Science , volume=

    Visualization and analysis of gene expression in tissue sections by spatial transcriptomics , author=. Science , volume=. 2016 , publisher=

    2016

  2. [2]

    Nature biomedical engineering , volume=

    Integrating spatial gene expression and breast tumour morphology via deep learning , author=. Nature biomedical engineering , volume=. 2020 , publisher=

    2020

  3. [3]

    BioRxiv , pages=

    Leveraging information in spatial transcriptomics to predict super-resolution gene expression from histology images in tumors , author=. BioRxiv , pages=. 2021 , publisher=

    2021

  4. [4]

    Briefings in Bioinformatics , volume=

    Spatial transcriptomics prediction from histology jointly through transformer and graph neural networks , author=. Briefings in Bioinformatics , volume=. 2022 , publisher=

    2022

  5. [5]

    Advances in Neural Information Processing Systems , volume=

    Spatially resolved gene expression prediction from histology images via bi-modal contrastive learning , author=. Advances in Neural Information Processing Systems , volume=

  6. [6]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

    Exemplar guided deep neural network for spatial transcriptomics analysis of gene expression prediction , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

  7. [7]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Accurate spatial gene expression prediction by integrating multi-resolution features , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  8. [8]

    Proceedings of the IEEE/CVF International Conference on computer vision , pages=

    SEPAL: spatial gene expression prediction from local graphs , author=. Proceedings of the IEEE/CVF International Conference on computer vision , pages=

  9. [9]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Merge: multi-faceted hierarchical graph-based gnn for gene expression prediction from whole slide histopathology images , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  10. [10]

    Briefings in Bioinformatics , volume=

    THItoGene: a deep learning method for predicting spatial transcriptomics from histological images , author=. Briefings in Bioinformatics , volume=. 2024 , publisher=

    2024

  11. [11]

    Pattern Recognition , volume=

    Spatial transcriptomics analysis of gene expression prediction using exemplar guided graph neural network , author=. Pattern Recognition , volume=. 2024 , publisher=

    2024

  12. [12]

    2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) , pages=

    Boundary-guided learning for gene expression prediction in spatial transcriptomics , author=. 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) , pages=. 2024 , organization=

    2024

  13. [13]

    Medical Image Analysis , pages=

    Completing spatial transcriptomics data for gene expression prediction benchmarking , author=. Medical Image Analysis , pages=. 2025 , publisher=

    2025

  14. [14]

    Nature Communications , volume=

    Combining spatial transcriptomics with tissue morphology , author=. Nature Communications , volume=. 2025 , publisher=

    2025

  15. [15]

    arXiv preprint arXiv:2604.03630 , year=

    A Multimodal Foundation Model of Spatial Transcriptomics and Histology for Biological Discovery and Clinical Prediction , author=. arXiv preprint arXiv:2604.03630 , year=

    Pith/arXiv arXiv

  16. [16]

    arXiv preprint arXiv:2411.16793 , year=

    St-align: A multimodal foundation model for image-gene alignment in spatial transcriptomics , author=. arXiv preprint arXiv:2411.16793 , year=

    Pith/arXiv arXiv

  17. [17]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Adapting a Pre-trained Single-Cell Foundation Model to Spatial Gene Expression Generation from Histology Images , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  18. [18]

    Nature , volume=

    Exploring tissue architecture using spatial transcriptomics , author=. Nature , volume=. 2021 , publisher=

    2021

  19. [19]

    Nature Reviews Genetics , volume=

    Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics , author=. Nature Reviews Genetics , volume=. 2021 , publisher=

    2021

  20. [20]

    Briefings in Bioinformatics , volume=

    Deep learning in spatially resolved transcriptomics: a comprehensive technical view , author=. Briefings in Bioinformatics , volume=. 2024 , publisher=

    2024

  21. [21]

    2025 , eprint=

    Spatial Transcriptomics Expression Prediction from Histopathology Based on Cross-Modal Mask Reconstruction and Contrastive Learning , author=. 2025 , eprint=

    2025

  22. [22]

    arXiv preprint arXiv:1609.02907 , year=

    Semi-supervised classification with graph convolutional networks , author=. arXiv preprint arXiv:1609.02907 , year=

    Pith/arXiv arXiv

  23. [23]

    arXiv preprint arXiv:1710.10903 , year=

    Graph attention networks , author=. arXiv preprint arXiv:1710.10903 , year=

    Pith/arXiv arXiv

  24. [24]

    Advances in neural information processing systems , volume=

    Inductive representation learning on large graphs , author=. Advances in neural information processing systems , volume=

  25. [25]

    International conference on machine learning , pages=

    Neural message passing for quantum chemistry , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  26. [26]

    arXiv preprint arXiv:1903.02428 , year=

    Fast graph representation learning with PyTorch Geometric , author=. arXiv preprint arXiv:1903.02428 , year=

    Pith/arXiv arXiv 1903

  27. [27]

    European semantic web conference , pages=

    Modeling relational data with graph convolutional networks , author=. European semantic web conference , pages=. 2018 , organization=

    2018

  28. [28]

    Nature methods , volume=

    SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network , author=. Nature methods , volume=. 2021 , publisher=

    2021

  29. [29]

    Nature communications , volume=

    Deciphering spatial domains from spatially resolved transcriptomics with an adaptive graph attention auto-encoder , author=. Nature communications , volume=. 2022 , publisher=

    2022

  30. [30]

    Nature communications , volume=

    Spatially informed clustering, integration, and deconvolution of spatial transcriptomics with GraphST , author=. Nature communications , volume=. 2023 , publisher=

    2023

  31. [31]

    NPB-REC: A non-parametric Bayesian deep-learning approach for undersampled MRI reconstruction with uncertainty estimation , volume=

    Khawaled, Samah and Freiman, Moti , year=. NPB-REC: A non-parametric Bayesian deep-learning approach for undersampled MRI reconstruction with uncertainty estimation , volume=. doi:10.1016/j.artmed.2024.102798 , journal=

    arXiv 2024

  32. [32]

    Artificial intelligence in medicine , volume=

    Stable feature selection utilizing Graph Convolutional Neural Network and Layer-wise Relevance Propagation for biomarker discovery in breast cancer , author=. Artificial intelligence in medicine , volume=. 2024 , publisher=

    2024

  33. [33]

    Methods , volume=

    Prediction of drug-induced hepatotoxicity based on histopathological whole slide images , author=. Methods , volume=. 2023 , publisher=

    2023

  34. [34]

    Frontiers in genetics , volume=

    Graph neural networks and their current applications in bioinformatics , author=. Frontiers in genetics , volume=. 2021 , publisher=

    2021

  35. [35]

    Nature Methods , volume=

    Benchmarking spatial clustering methods with spatially resolved transcriptomics data , author=. Nature Methods , volume=. 2024 , publisher=

    2024

  36. [36]

    Computational and Structural Biotechnology Journal , volume=

    A comprehensive overview of graph neural network-based approaches to clustering for spatial transcriptomics , author=. Computational and Structural Biotechnology Journal , volume=. 2023 , publisher=

    2023

  37. [37]

    Nature Computational Science , volume=

    Cell clustering for spatial transcriptomics data with graph neural networks , author=. Nature Computational Science , volume=. 2022 , publisher=

    2022

  38. [38]

    Advanced Science , volume=

    SpaBatch: Deep Learning-Based Cross-Slice Integration and 3D Spatial Domain Identification in Spatial Transcriptomics , author=. Advanced Science , volume=. 2025 , publisher=

    2025

  39. [39]

    Advances in neural information processing systems , volume=

    Deep evidential regression , author=. Advances in neural information processing systems , volume=

  40. [40]

    international conference on machine learning , pages=

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning , author=. international conference on machine learning , pages=. 2016 , organization=

    2016

  41. [41]

    Advances in neural information processing systems , volume=

    Simple and scalable predictive uncertainty estimation using deep ensembles , author=. Advances in neural information processing systems , volume=

  42. [42]

    arXiv preprint arXiv: 2107.07511 , year=

    A gentle introduction to conformal prediction and distribution-free uncertainty quantification , author=. arXiv preprint arXiv: 2107.07511 , year=

    Pith/arXiv arXiv

  43. [43]

    International conference on machine learning , pages=

    Accurate uncertainties for deep learning using calibrated regression , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

  44. [44]

    Artificial Intelligence in Medicine , volume=

    Trustworthy clinical AI solutions: a unified review of uncertainty quantification in deep learning models for medical image analysis , author=. Artificial Intelligence in Medicine , volume=. 2024 , publisher=

    2024

  45. [45]

    Nature communications , volume=

    Uncertainty-informed deep learning models enable high-confidence predictions for digital histopathology , author=. Nature communications , volume=. 2022 , publisher=

    2022

  46. [46]

    Advances in neural information processing systems , volume=

    Evidential deep learning to quantify classification uncertainty , author=. Advances in neural information processing systems , volume=

  47. [47]

    Genome Biology , volume=

    Evaluating spatially variable gene detection methods for spatial transcriptomics data , author=. Genome Biology , volume=. 2024 , publisher=

    2024

  48. [48]

    BioRxiv , year=

    Benchmarking computational methods to identify spatially variable genes and peaks , author=. BioRxiv , year=

  49. [49]

    Nature methods , volume=

    SpatialDE: identification of spatially variable genes , author=. Nature methods , volume=. 2018 , publisher=

    2018

  50. [50]

    Nature medicine , volume=

    Towards a general-purpose foundation model for computational pathology , author=. Nature medicine , volume=. 2024 , publisher=

    2024

  51. [51]

    Nature , volume=

    A whole-slide foundation model for digital pathology from real-world data , author=. Nature , volume=. 2024 , publisher=

    2024

  52. [52]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  53. [53]

    arXiv preprint arXiv:2010.11929 , year=

    An image is worth 16x16 words: Transformers for image recognition at scale , author=. arXiv preprint arXiv:2010.11929 , year=

    Pith/arXiv arXiv 2010

  54. [54]

    Nucleic acids research , volume=

    Enrichr: a comprehensive gene set enrichment analysis web server 2016 update , author=. Nucleic acids research , volume=. 2016 , publisher=

    2016

  55. [55]

    arXiv preprint arXiv:1802.03426 , year=

    Umap: Uniform manifold approximation and projection for dimension reduction , author=. arXiv preprint arXiv:1802.03426 , year=

    Pith/arXiv arXiv

  56. [56]

    Psychometrika , volume=

    Note on the sampling error of the difference between correlated proportions or percentages , author=. Psychometrika , volume=. 1947 , publisher=

    1947

  57. [57]

    Journal of the Royal statistical society: series B (Methodological) , volume=

    Controlling the false discovery rate: a practical and powerful approach to multiple testing , author=. Journal of the Royal statistical society: series B (Methodological) , volume=. 1995 , publisher=

    1995

  58. [58]

    Monographs on statistics and applied probability , volume=

    An introduction to the bootstrap , author=. Monographs on statistics and applied probability , volume=

  59. [59]

    2013 , publisher=

    Statistical power analysis for the behavioral sciences , author=. 2013 , publisher=

    2013

  60. [60]

    Encyclopedia of biostatistics , volume=

    Spearman rank correlation , author=. Encyclopedia of biostatistics , volume=. 2005 , publisher=

    2005

  61. [61]

    Noise reduction in speech processing , pages=

    Pearson correlation coefficient , author=. Noise reduction in speech processing , pages=. 2009 , publisher=

    2009

  62. [62]

    R package Kendall , volume=

    Kendall rank correlation and Mann-Kendall trend test , author=. R package Kendall , volume=

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.