SUICA: Learning Super-high Dimensional Sparse Implicit Neural Representations for Spatial Transcriptomics

Jun Ding; Qingtian Zhu; Yifan Zhan; Yinqiang Zheng; Yuling Sang; Yumin Zheng; Ziyan Zhu

REVIEW 3 major objections 4 minor 1 cited by

SUICA: Learning Super-high Dimensional Sparse Implicit Neural Representations for Spatial Transcriptomics

T0 review · 3 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Sparse gene-expression data can be modeled as a continuous neural map that reconstructs unmeasured tissue spots with better fidelity and biological structure than existing tools.

desk verdict SUICA is a plausible GAE+INR recipe for spatial transcriptomics, but the central spatial-imputation benchmark may be transductive because the graph autoencoder appears to see the test spots before evaluation. read the letter →

arxiv 2412.01124 v2 pith:ICGG5LFG submitted 2024-12-02 cs.LG q-bio.GN

Qingtian Zhu , Yumin Zheng , Yuling Sang , Yifan Zhan , Ziyan Zhu , Jun Ding , Yinqiang Zheng This is my paper · ORCID

classification cs.LGq-bio.GN

keywords spatialtranscriptomicsimplicitneuralrepresentationgraphautoencoderimputationgenedenoisingzero-inflateddistributionregression-by-classification

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

SUICA aims to model spatial transcriptomics (ST) data — sparse measurements of tens of thousands of genes at discrete, irregularly placed tissue spots — as a continuous, compact function of spatial coordinates. The central claim is that a graph-aware autoencoder coupled with an implicit neural representation (INR), trained under a classification-style loss, lets one pipeline reconstruct unmeasured spots, impute muted genes, and denoise counts across several ST platforms without knowing the degradation type. The stakes are practical: ST data are expensive and dropout-heavy, so a faithful continuous representation could let researchers densify and clean measured slices for downstream analysis. On the main benchmark, the paper reports the lowest errors, cosine similarity 5.6% above the runner-up, Pearson and Spearman correlations of 0.792 and 0.447, and cell-type clustering (ARI 0.343) above the ground-truth slice's reference (0.312).

What carries the argument

The load-bearing machinery is the GAE–INR–decoder cascade. An implicit neural representation (INR) is a neural network that maps spatial coordinates to signal values, giving a continuous model of a discretely sampled signal; SUICA's INR learns $\mathbf{x} \mapsto \hat{z}$ from spot coordinates to the low-dimensional embeddings $z$ produced by the graph autoencoder. The GAE, built on a graph convolutional network in its encoder only, supplies embeddings that are compact, dense, and context-aware; the decoder is detached from the encoder, warmed up separately, then finetuned with the INR frozen to map $\hat{z}$ back to the raw expression space. The regression-by-classification term — a Dice loss computed on $\tanh$-mapped outputs — is what stops the reconstruction from collapsing the zero-inflated distribution into a smooth Gaussian one, and the choice of FFN versus SIREN as the INR backbone is made according to spatial density of the slice.

What would settle it

Re-run the spatial-imputation protocol with the GAE pre-trained strictly on the 80% training spots and the KNN graph built without the held-out 20%, then compare MAE, MSE, cosine similarity, Pearson, Spearman, and ARI against the reported MOSTA values of 5.66, 0.85, 0.797, 0.792, 0.447, and 0.343. If the margins over FFN, SIREN, and STAGE shrink substantially, or if the ARI no longer clears the ground-truth reference of 0.312, the spatial-generalization claim would be falsified.

Watch

Extended reading notes

Core claim

The paper's discovery is that the real obstacle to applying INRs to ST is not network capacity but the geometry of the target space: raw expression profiles are zero-inflated and can exceed 20,000 channels, so an INR regressing directly onto them washes out the sparsity. SUICA factors the problem: a graph-augmented autoencoder compresses the raw cell-by-gene matrix into low-dimensional, structure-aware embeddings, the INR learns the smooth coordinate-to-embedding map, and a finetuned decoder returns to raw expression space. The skewed zero-versus-nonzero balance is treated as a quasi-classification problem: outputs are pushed through a non-negative tanh and scored with a Dice (intersection-over-union) loss that preserves sparsity in the predictions. The paper presents graph total variation evidence that GAE embeddings retain sharper spatial structure than plain autoencoder embeddings, and reports that the full pipeline beats vanilla INR variants and imputation baselines on numerical fidelity, statistical correlation, and bio-conservation while also amplifying biological signatures for downstream analysis.

Load-bearing premise

The load-bearing premise is that the held-out 20% of spots were genuinely unseen during graph-autoencoder pre-training; the paper (Sections 3.2.1 and 4.2) never states whether the GAE was fit on only the training subset or on the full slice, so the spatial-imputation benchmark may be autoencoding already-seen data rather than testing spatial generalization.

Editorial extensions

If this is right

Spatial imputation: expression profiles of held-out spots can be predicted from coordinates alone, with the reported benchmark showing the lowest MAE/MSE, the highest cosine similarity (5.6% above the runner-up), and the best Pearson/Spearman correlations among FFN, SIREN, and STAGE.
Gene imputation and denoising use the identical pipeline with no knowledge of the degradation, making SUICA a reference-free, degradation-agnostic restoration tool rather than a per-task model.
Predicted profiles preserve cell-type structure: clustering SUICA's reconstructions reaches an ARI of 0.343 on the MOSTA slice, above the 0.312 reference ARI of the ground truth, and marker genes such as RPLP0 and AFP localize to the correct anatomical regions.
A single continuous representation spans Stereo-seq, Slide-seqV2, Visium, and MERFISH, so the method is claimed to transfer across sequencing platforms with different spatial densities and dropout rates.

Reading between the lines

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

The decisive check on the spatial-generalization claim is to pre-train the GAE strictly on the 80% training spots, with the KNN graph built without the held-out 20%; the paper's own wording leaves open whether the GAE saw the test spots during pre-training.
The decoupling recipe — let a smooth neural field do spatial interpolation in a learned latent space while a separate head enforces raw-space fidelity — transfers naturally to other sparse, high-dimensional coordinate measurements, such as single-cell atlases or multi-modal tissue imaging, where direct INR regression suffers the same zero-inflation failure.
The ARI above the ground-truth reference implies the representation acts as a denoiser that sharpens cell-type boundaries; a testable prediction is that the amplification concentrates along known anatomical or developmental gradients, and a caveat is that low-signal regions could be over-sharpened.

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.

Desk Editor's Note

read the letter

The thing to know about this paper: the method is reasonable but the main evaluation may not test what it claims. SUICA combines a graph autoencoder, an INR over spatial coordinates, and a Dice-loss regularizer to reconstruct gene expression. The combination is new in the spatial transcriptomics space, the code is released, and the authors evaluate across Stereo-seq, Slide-seqV2, Visium, and MERFISH. The ablation study (AE, Dice, Graph) is informative, and the observation that SIREN versus FFN matters depending on spatial density is a useful practical detail.

The soft spot is load-bearing. The spatial imputation protocol splits spots 80/20, but Section 3.2.1 says the GAE is pre-trained on “the given ST slice” and that zgt is obtained for “all spots.” Section 3.2.4 then trains the decoder to fit ygt without restricting to the training subset. If the GAE encoder, GAE decoder, and decoder fine-tuning all see the 20% test spots’ expression profiles, then Table 1 measures transductive autoencoding plus coordinate interpolation, not spatial generalization to unseen spots. The paper never explicitly rules this out, and Appendix F’s protocol (varying training fraction while test stays 20%) makes it plausible that unused spots are still fed to the GAE. This needs to be stated plainly and, if the GAE does see all spots, the benchmark needs to be redone with the GAE trained only on training spots.

Other issues are real but less severe. Numerical fidelity is reported only on non-zero entries, which flatters zero-inflated data; no standard deviations or repeated seeds are given; and SUICA’s ARI on Stereo-seq exceeds the ground-truth ARI, which is a red flag that the method is doing something beyond faithful reconstruction. The novelty claim is also softened by the fact that the INR maps coordinates to a 32-d embedding, not directly to 20,000 genes. Table 7 on MERFISH shows SUICA’s Spearman is actually the worst of the compared methods, so the “superior across platforms” line should be moderated.

The arXiv v2 text itself is a corrupted compilation: it contains duplicated sections, leftover “CVPR#1357 CONFIDENTIAL REVIEW COPY” headers, and an author note (“[Qingtian: check Prof. Ding’s version]”). That is not the authors’ science, but it makes the manuscript hard to rely on as-is and must be cleaned up.

Who is this for? Someone working on ST imputation or INR applications to sparse high-dimensional data. The method has a good chance of being useful for gene imputation and denoising, which are evaluated on all spots with synthetic degradation and are not undermined by the transductive concern. The spatial imputation claim urgently needs the protocol clarified. I would send this to review, but with a request to the authors to specify the GAE training split, rerun the benchmark under a clean held-out protocol, report variance across seeds, and soften the cross-platform claim. If the transductive concern is confirmed, the spatial imputation results in Table 1 cannot stand as evidence of spatial generalization.

Referee Report

3 major / 4 minor

Summary. The paper proposes SUICA, an implicit neural representation (INR) method for spatial transcriptomics (ST). A graph autoencoder (GAE) first maps the high-dimensional, sparse expression matrix into low-dimensional embeddings; an INR then maps spatial coordinates to those embeddings; a finetuned decoder maps the interpolated embeddings back to the raw expression space, using a Dice-loss regularizer to handle zero inflation. The authors evaluate SUICA on spatial imputation, gene imputation, and denoising across Stereo-seq MOSTA, Slide-seqV2, Visium, and MERFISH datasets, reporting improvements in MAE, MSE, cosine similarity, Pearson/Spearman correlations, and ARI over INR baselines and STAGE, with code made available.

Significance. If the reported results are valid, SUICA is a meaningful contribution: it offers a modular INR design that scales to the super-high-dimensional, zero-inflated output space of ST, and the paper provides multi-platform experiments, ablations, and a data-efficiency study in the appendix. The release of code is a concrete strength. However, the central spatial-imputation benchmark rests on an evaluation protocol that is not fully specified and may be circular, because the GAE appears to be pretrained on the full slice including the held-out test spots. Until that issue is resolved, the quantitative claims in Table 1 cannot be taken as evidence of generalization to unseen spatial locations.

major comments (3)

[§3.2.1, §3.2.2, §3.2.4, §4.2, Appendix F] The spatial-imputation evaluation may be transductive. Section 3.2.1 says the spots are split into training and test subsets and that SUICA is trained on the training subset, but Section 3.2.2 says the GAE is pretrained "using the given ST slice in a self-regressing manner" and that zgt is obtained "for all spots," and Section 3.2.4 says the decoder is trained to fit the "raw dataset readouts ygt" without restricting to the training subset. If the GAE encoder, GAE decoder, the latent codes zgt for the 20% test spots, and the decoder finetuning all use the test spots' expression profiles, then Table 1 measures transductive autoencoding plus coordinate interpolation rather than spatial imputation of unseen spots. This would invalidate the central claim in Section 4.4. The authors must state unambiguously how the held-out 20% is excluded from every stage—GAE pretraining, latent-code computation, INR fitting, and decoder finetuning—and rerun the benchmark under that protocol. The same ambiguity affects the gene-imputation results in Table 2, because the GAE is pretrained on the unmuted expression matrix before the 70% masking step.
[§4.1, Table 1] Numerical fidelity is reported only on non-zero entries. For zero-inflated ST data, a predictor that always outputs zero achieves zero error on the zero entries and is never penalized for false positives under this protocol, so the MAE/MSE/cosine numbers do not establish fidelity for the full expression profile. The paper should report full-entry MSE/MAE and a zero/non-zero classification metric (e.g., IoU or F1) alongside the non-zero-only metrics, or justify with a sensitivity analysis why the non-zero-only choice cannot change the conclusions.
[§4.2–§4.4, Tables 1–5] No measure of variability is reported. All comparisons are single numbers without standard deviations, number of random splits, or seeds, and several claimed margins are small (e.g., the 4.5–5.6% improvements reported in Section 4.4). Without repeated trials or confidence intervals, the observed differences may not be statistically significant. Report mean and standard deviation over at least three random 80/20 splits or training seeds for the main tables.

minor comments (4)

[Throughout] The manuscript contains duplicated blocks of text and CVPR submission boilerplate (e.g., repeated paragraphs around Section 3.2.4 and the phrase "CVPR#1357 CONFIDENTIAL REVIEW COPY"). These should be removed before resubmission.
[§3.2.4, Eqs. (2)–(5)] The loss definitions are inconsistent across versions: one formulation of Dice loss uses sgn(ygt) in the numerator, another uses ygt directly, and Eq. (5) reads "LMAE = LMSE," which appears to be a typo. Please harmonize the equations and define all symbols precisely.
[Appendix F] The data-efficiency ablation varies the training fraction while keeping the test set at 20%, but the text does not state whether the unused spots are discarded or are still used for GAE pretraining. This directly affects the interpretation of the results and should be clarified.
[§4.4] The claim that SUICA's ARI exceeds the ground-truth ARI by 3.9% is surprising and should be discussed explicitly: if the predicted data are more clusterable than the ground truth, the paper should explain whether this reflects desirable denoising, an artifact of the metric, or a mismatch with biological cell-type structure.

Circularity Check

1 steps flagged · score 6.0 of 10

Spatial imputation benchmark may be transductive: the GAE is pretrained on the whole ST slice and obtains zgt for all spots, so the 20% held-out spots' expression profiles enter the model before evaluation; Table 1 may not test unseen-spot prediction.

fitted input called prediction [Section 3.2.1 (Overview), Section 3.2.4 (Decoding Head), Section 4.3/4.4 (Implemented Details and Spatial Imputation)]
"we first incorporate a graph-based encoder and pre-train Graph Autoencoder (GAE) using the given ST slice in a self-regressing manner. With the pre-trained GAE, we obtain the encoded latent representation for all spots, denoted as zgt. ... After we have pre-trained a GAE, we use its encoder to generate the embeddings for all of the spots, which we denote as zgt."

The spatial-imputation evaluation is defined over an 80/20 train/test split of spots, and Section 4.4 reports the Table 1 results as predicting gene expression of 'unseen spots'. But the GAE pretraining is described as using the entire ST slice, and zgt is generated for all spots, including the 20% held-out spots. The GAE reconstruction loss (Eq. 1) and the decoder finetuning loss (Eq. 4) are written over ygt without any restriction to the 80% subset. Hence the expression profiles of the 'test' spots enter the encoder that produces zgt and the decoder that maps z back to y.

full rationale

Aside from the spatial-imputation evaluation leak, I do not find other circularity. The Dice-loss regression-by-classification scheme, the INR smoothness prior, and the GAE embedding choice are independent modeling decisions with separate external benchmarks and do not reduce to their inputs. There is no load-bearing self-citation chain and no imported uniqueness theorem. The central derivation is therefore self-contained except that the headline 'unseen spot' claim is compromised by the all-spot GAE pretraining and all-spot zgt construction described in Sections 3.2.1 and 3.2.4; Appendix F's wording that the test set 'remains the same (20% of all spots)' while the training fraction varies over the whole dataset is consistent with the whole slice being available during pretraining. If the GAE and decoder finetuning were restricted to the 80% training subset, the circularity would disappear. Score 6 reflects that, under the paper's own protocol description, the main spatial-imputation prediction reduces in part to reconstruction of already-seen expression profiles.

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

The central claim rests on standard approximation theory, the domain assumption that ST signals are spatially smooth and zero-inflated, and the biological informativeness of spatial KNN graphs. There are no invented entities. The main free parameters are the latent dimension, the Dice loss weight (not reported), the INR backbone frequencies, and the graph size.

free parameters (5)

embedding dimension = 32
The dimension of the GAE latent space is set to 32 by hand; the main INR output dimension and thus the complexity of the spatial mapping depend on this choice.
Dice loss weight lambda = not reported
Equation (4) includes lambda to balance Dice Loss against MSE and MAE; the value is never given, so the final objective is not fully specified.
SIREN angular frequency omega = 30
Appendix B states omega=30 is the default; it controls the frequency of the periodic activations and is chosen by hand.
FFN random Fourier feature mapping size = 256
Appendix B states the mapping size is 256; it affects the capacity of the FFN backbone.
KNN neighbors k for GAE graph = 5
The graph for GCN uses 5 nearest neighbors including the cell itself; this choice controls the spatial context propagated by the GCN.

assumptions (4)

standard math Universal approximation: MLPs can approximate the coordinate-to-expression mapping given enough capacity.
The paper invokes INR approximation capability in the abstract and Section 2.1; it relies on the universal approximation theorems of Cybenko and Hornik.
domain assumption Gene expression is a smooth function of spatial coordinates; the continuity and smoothness prior of INRs yields meaningful interpolation between spots.
This is the core modeling assumption stated in the introduction and used in all imputation tasks.
domain assumption Spatial KNN graphs capture biologically relevant cellular context; the GCN's message passing on this graph improves embeddings.
This motivates the graph-augmented autoencoder in Section 3.2.2; if the graph structure is not biologically informative, the GAE gain would vanish.
domain assumption Zero inflation in ST data is a degradation to be corrected, and the non-zero measurements are the truth to preserve.
This assumption underlies the non-zero-only evaluation and the Dice-loss objective in Sections 3.2.4 and 4.1.

how reviews work

0 comments

Cite this review

Pith. "Pith review of SUICA: Learning Super-high Dimensional Sparse Implicit Neural Representations for Spatial Transcriptomics." pith.science (2026). https://pith.science/paper/ICGG5LFG

@misc{pith2026241201124,
  author       = {Pith},
  title        = {Pith review of: SUICA: Learning Super-high Dimensional Sparse Implicit Neural Representations for Spatial Transcriptomics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ICGG5LFG}},
  note         = {Machine review of arXiv:2412.01124}
}

read the original abstract

Spatial Transcriptomics (ST) is a method that captures gene expression profiles aligned with spatial coordinates. The discrete spatial distribution and the super-high dimensional sequencing results make ST data challenging to be modeled effectively. In this paper, we manage to model ST in a continuous and compact manner by the proposed tool, SUICA, empowered by the great approximation capability of Implicit Neural Representations (INRs) that can enhance both the spatial density and the gene expression. Concretely within the proposed SUICA, we incorporate a graph-augmented Autoencoder to effectively model the context information of the unstructured spots and provide informative embeddings that are structure-aware for spatial mapping. We also tackle the extremely skewed distribution in a regression-by-classification fashion and enforce classification-based loss functions for the optimization of SUICA. By extensive experiments of a wide range of common ST platforms under varying degradations, SUICA outperforms both conventional INR variants and SOTA methods regarding numerical fidelity, statistical correlation, and bio-conservation. The prediction by SUICA also showcases amplified gene signatures that enriches the bio-conservation of the raw data and benefits subsequent analysis. The code is available at https://github.com/Szym29/SUICA.

Figures

Figures reproduced from arXiv: 2412.01124 by the authors.

**Figure 1.** Starting with the discretely sampled spots (a) of ST, SUICA performs continuous modeling (b) by aid of the great approximation power of INRs. This approach enables complete profiling of cell heterogeneity, as visualized in the UMAP (c), which further facilitating the discovery of new biology. 1. Introduction Spatial Transcriptomics (ST) enables scientists to quantify gene expression while preserving spatial informat… view at source ↗

**Figure 2.** The overall pipeline of SUICA. At training-time, a GAE based on cell graphs is trained, with whose pre-trained decoder concatenated to a INR. The INR then maps spot coordinates to the corresponding gene expressions. SUICA is capable of performing spatial imputation, gene imputation, and denoising. et al., 2021; Oechsle et al., 2021). As for video compression, NeRV (Chen et al., 2021) and subsequent variants (Li et … view at source ↗

**Figure 1.** Starting with the discretely sampled spots (a) of ST, leveraging cell relationships hav [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

Figures from the paper (5 more)

**Figure 3.** Figure 3: Spectral analysis with the embeddings of AE and GAE. GAE yields structure-aware and disentangled embeddings with high-frequency details. GTV: Graph Total Variation. is integrated through a graph structure, which we model using a Graph Convolutional Network (GCN). Impor…

**Figure 4.** Figure 4: Visual comparisons of predicted marker gene expressions on MOSTA dataset (Chen et al., 2022a) and Slide-seqV2 Mouse hippocampus dataset (Stickels et al., 2021), with the descriptions of the markers attached below the results. based reference-free baseline model STAGE (…

**Figure 5.** Figure 5: Spatially visualized comparison on bio-conservations of predicted spots on MOSTA mouse embryo E16.5, Slide-seqV2 mouse hippocampus, and Visium-Mouse brain. to the ground-truth data (Luecken et al., 2022). A high bioconservation score indicates that the predicted data …

**Figure 6.** Figure 6: Workflow of sequencing-based spatial transcriptomics sequencing. (a) Spatially indexing, with microarrayer for 10x Visium, nanobead technology for Slide-seqV2 and nanoball-based strategy for Stereo-seq; (b) Placing the slice of tissue on the indexed surface and captu…

**Figure 7.** Figure 7: Spatially visualized predicted spots on MERFISH human heart (Farah et al., 2024) [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

Continuous Representation Methods, Theories, and Applications: An Overview and Perspectives
cs.CV 2025-05 conditional novelty 2.0 of 10

A survey organizing continuous representation methods into parametric models, structural modeling, theory, and applications, with a curated open-source reference repository.

Reference graph

Works this paper leans on

13 extracted references · 12 canonical work pages · cited by 1 Pith paper

[3]

Towards deeper un- derstanding of variational autoencoding models

Zhao, S., Song, J., and Ermon, S. Towards deeper un- derstanding of variational autoencoding models. arXiv preprint arXiv:1702.08658,

arXiv
[4]

MSE: ×10−2

Quantitative results of SUICA empowered by different INR backbones. MSE: ×10−2. Backbones Embryo E9.5 Embryo E16.5MSE↓ Cosine↑ Pearson↑ MSE↓ Cosine↑ Pearson↑ FFN (Tancik et al., 2020)0.55 0.752 0.799 1.47 0.807 0.761SIREN (Sitzmann et al., 2020)0.52 0.766 0.814 1.62 0.786 0.745 Table

work page 2020
[5]

MAE/MSE: ×10−2

Quantitative comparison on Stereo-seq MOSTA dataset (Chen et al., 2022a) between SUICA and NGP (M ¨uller et al., 2022). MAE/MSE: ×10−2. Methods MAE↓ MSE↓ Cosine↑ Pearson↑ Spearman↑ NGP (M¨uller et al., 2022)6.39 1.14 0.729 0.742 0.415 SUICA 5.66 0.85 0.797 0.792 0.447 Explicit Alternatives Recently, researchers are employ- ing explicit or hybrid alternati...

work page 2022
[6]

Note that for Visium-Human Brain, there is no annotation of cell type for the evaluation of ARI, while reference ARI is 0.428 for Mouse Brain

Quantitative benchmarking results of spatial imputation on two Visium ST cases (Palla et al., 2022; Wei et al., 2022). Note that for Visium-Human Brain, there is no annotation of cell type for the evaluation of ARI, while reference ARI is 0.428 for Mouse Brain. Bold figures are best scores and underlined figures are second-best. MAE/MSE: ×10−2 for Human B...

work page 2022
[7]

These issues might lead to ambiguity in gene expression that is not wanted

and struggle to match AEs in re- construction accuracy (Dai et al., 2020). These issues might lead to ambiguity in gene expression that is not wanted. These limitations stem from the inherent trade-off between the quality of latent representation learning and reconstruc- tion fidelity in V AEs. Although variants like beta-V AE (Hig- gins et al.,

work page 2020
[8]

% rep- resents the proportion of the spots used for training while test set remains 20%

Ablation study on the data-efficiency of SUICA. % rep- resents the proportion of the spots used for training while test set remains 20%. MAE/MSE: ×10−2. % MAE↓ MSE↓ Cosine↑ Pearson↑ 80% 8.01 1.47 0.807 0.761 60% 7.96 1.52 0.801 0.752 40% 8.00 1.59 0.790 0.739 20% 8.14 1.62 0.786 0.738 strictly bounded, where the linear layer is usually clamped or compress...

work page 2021
[11]

Spatially visualized predicted spots on MERFISH human heart (Farah et al., 2024). Table

work page 2024
[12]

MAE/MSE: ×10−1

Quantitative results of spatial imputation on MERFISH human heart (Farah et al., 2024). MAE/MSE: ×10−1. Methods MAE↓ MSE↓ Cosine↑ Pearson↑ Spearman↑ FFN (Tancik et al., 2020)6.06 5.86 0.840 0.717 0.554SIREN (Sitzmann et al., 2020)5.54 4.98 0.864 0.759 0.606STAGE (Li et al., 2024)5.48 5.10 0.870 0.709 0.558SUICA 4.65 3.92 0.892 0.718 0.548 annotate the cel...

work page 2024

Show all 13 references

[2017]

Convolutional Occupancy Networks

Peng, S., Niemeyer, M., Mescheder, L., Pollefeys, M., and Geiger, A. Convolutional Occupancy Networks. In Com- puter Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part III 16, pp. 523–540. Springer,

2020
[2020]

5.76 0.881 0.772 0.786 0.402 5.95 5.85 0.832 0.741 0.581 0.000587 SIREN (Sitzmann et al., 2020)6.58 0.933 0.756 0.747 0.196 5.35 4.29 0.878 0.804 0.647 0.359 STAGE (Li et al.,

2020
[2021]

H., Li, W., Vercauteren, T., Ourselin, S., and Jorge Cardoso, M

Sudre, C. H., Li, W., Vercauteren, T., Ourselin, S., and Jorge Cardoso, M. Generalised dice overlap as a deep learning loss function for highly unbalanced segmen- tations. In Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support: Third I...

2017
[2023]

11 SUICA: Learning Super-high Dimensional Sparse Implicit Neural Representations for Spatial Transcriptomics A. Background Spatial Transcriptomics (ST) is a technique that represent- ing the gene expression at individual spots within a slice of sample, providing a detailed pic...

2023
[2024]

The progressive training paradigm also guarantees that each module is doing the assigned job, keeping the coupling at minimum

6.19 0.805 0.795 0.772 0.223 4.55 3.20 0.918 0.825 0.666 0.140 TRIPLEX (Chung et al., 2024)4.75 0.560 0.881 0.850 0.319 9.35 14.0 0.00 -0.00682 -0.00715 0.358 UNIv2 (Chen et al., 2024)7.30 1.41 0.723 0.633 0.129 6.94 7.88 0.790 0.631 0.425 0.228 SUICA(Ours) 4.99 0.567 0.860 0....

2024

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [3]

Towards deeper un- derstanding of variational autoencoding models

Zhao, S., Song, J., and Ermon, S. Towards deeper un- derstanding of variational autoencoding models. arXiv preprint arXiv:1702.08658,

arXiv

[2] [4]

MSE: ×10−2

Quantitative results of SUICA empowered by different INR backbones. MSE: ×10−2. Backbones Embryo E9.5 Embryo E16.5MSE↓ Cosine↑ Pearson↑ MSE↓ Cosine↑ Pearson↑ FFN (Tancik et al., 2020)0.55 0.752 0.799 1.47 0.807 0.761SIREN (Sitzmann et al., 2020)0.52 0.766 0.814 1.62 0.786 0.745 Table

work page 2020

[3] [5]

MAE/MSE: ×10−2

Quantitative comparison on Stereo-seq MOSTA dataset (Chen et al., 2022a) between SUICA and NGP (M ¨uller et al., 2022). MAE/MSE: ×10−2. Methods MAE↓ MSE↓ Cosine↑ Pearson↑ Spearman↑ NGP (M¨uller et al., 2022)6.39 1.14 0.729 0.742 0.415 SUICA 5.66 0.85 0.797 0.792 0.447 Explicit Alternatives Recently, researchers are employ- ing explicit or hybrid alternati...

work page 2022

[4] [6]

Note that for Visium-Human Brain, there is no annotation of cell type for the evaluation of ARI, while reference ARI is 0.428 for Mouse Brain

Quantitative benchmarking results of spatial imputation on two Visium ST cases (Palla et al., 2022; Wei et al., 2022). Note that for Visium-Human Brain, there is no annotation of cell type for the evaluation of ARI, while reference ARI is 0.428 for Mouse Brain. Bold figures are best scores and underlined figures are second-best. MAE/MSE: ×10−2 for Human B...

work page 2022

[5] [7]

These issues might lead to ambiguity in gene expression that is not wanted

and struggle to match AEs in re- construction accuracy (Dai et al., 2020). These issues might lead to ambiguity in gene expression that is not wanted. These limitations stem from the inherent trade-off between the quality of latent representation learning and reconstruc- tion fidelity in V AEs. Although variants like beta-V AE (Hig- gins et al.,

work page 2020

[6] [8]

% rep- resents the proportion of the spots used for training while test set remains 20%

Ablation study on the data-efficiency of SUICA. % rep- resents the proportion of the spots used for training while test set remains 20%. MAE/MSE: ×10−2. % MAE↓ MSE↓ Cosine↑ Pearson↑ 80% 8.01 1.47 0.807 0.761 60% 7.96 1.52 0.801 0.752 40% 8.00 1.59 0.790 0.739 20% 8.14 1.62 0.786 0.738 strictly bounded, where the linear layer is usually clamped or compress...

work page 2021

[7] [11]

Spatially visualized predicted spots on MERFISH human heart (Farah et al., 2024). Table

work page 2024

[8] [12]

MAE/MSE: ×10−1

Quantitative results of spatial imputation on MERFISH human heart (Farah et al., 2024). MAE/MSE: ×10−1. Methods MAE↓ MSE↓ Cosine↑ Pearson↑ Spearman↑ FFN (Tancik et al., 2020)6.06 5.86 0.840 0.717 0.554SIREN (Sitzmann et al., 2020)5.54 4.98 0.864 0.759 0.606STAGE (Li et al., 2024)5.48 5.10 0.870 0.709 0.558SUICA 4.65 3.92 0.892 0.718 0.548 annotate the cel...

work page 2024