Pith. sign in

REVIEW 5 major objections 6 minor 56 references

Brain-wide interpolation and conditioning of gene expression in the human brain using Implicit Neural Representations

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper argues that implicit neural representations, conditioned on a spectral-embedding gene axis, can convert the Allen Human Brain Atlas's sparse samples into reliable voxel-level expression maps across the whole human brain for any…

desk verdict A competent proof-of-concept for INR-based gene expression interpolation with a genuinely new gene-axis conditioning idea, but the voxel-level accuracy claim is unvalidated because both comparisons are against region-level interpolants of the same sparse data. read the letter →

arxiv 2506.11158 v1 pith:N56ARMAD submitted 2025-06-11 q-bio.GN cs.LG

Xizheng Yu , Justin Torok , Sneha Pandya , Sourav Pal , Vikas Singh , Ashish Raj This is my paper
classification q-bio.GNcs.LG
keywords implicitneuralrepresentationsspatialtranscriptomicsgeneexpressioninterpolationAllenHumanBrainAtlasAlzheimer'sdiseasetau-PETspectralembeddingSIREN
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

This paper claims that a type of neural network called an implicit neural representation (INR), aided by a gene-ordering axis from spectral embedding, can interpolate the Allen Human Brain Atlas's sparse microarray samples into continuous, voxel-level gene-expression maps across the whole human brain. The maps are produced for 100 Alzheimer's disease risk genes, and although no ground truth exists at unsampled locations, the authors validate indirectly: regional averages of their maps agree with the standard abagen interpolation (R=0.70), and the maps reproduce abagen's gene-wise correlations with tau-PET pathology in ADNI patients (R=0.89), including canonical risk genes like APOE and MAPT. If correct, this gives researchers queryable whole-brain expression maps at a spatial resolution the raw data cannot support, without additional tissue sampling. The approach is framed as disease-agnostic, with the AD gene set serving both as a use case and as a constraint.

What carries the argument

The central object is the SIREN implicit neural representation: a 12-layer MLP with 512 hidden units and sinusoidal activations that is trained on input tuples $(x,y,z,c,\gamma(e_g))$ and outputs normalized expression $v$, thereby defining an implicit continuous function that can be queried at arbitrary brain coordinates. The conditioning input carries the argument: $c$ encodes grey-matter (1), white-matter (−1), or neither (0), and $\gamma(e_g)$ is a NeRF-style positional encoding of a one-dimensional spectral-embedding score $e_g$ per gene. That score is the first non-trivial eigenvector of the normalized Laplacian of the 100-gene correlation matrix; it orders genes along an axis on which spatially similar genes are adjacent, and it is the mechanism that allows the network to share statistical strength across genes to fight spatial undersampling. The bandwidth-limited sinusoidal composition supplies the smoothing prior the authors rely on to fill unsampled voxels.

What would settle it

Measure expression directly at a set of unsampled coordinates chosen for sharp transcriptional transitions — cortical layer boundaries, hippocampal subfields, grey/white-matter borders — using an independent technique such as in situ hybridization or spatial RNA-sequencing, and compare those point measurements with the INR's predictions at the same MNI coordinates; if point-wise correlation falls far below the regional R≈0.67–0.70 or errors concentrate at the boundaries, the smoothness prior is false. A cheaper version is to hold out entire tissue samples or whole anatomical parcels during training and test whether the model predicts them at their true coordinates.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that a SIREN-based INR, conditioned on a one-dimensional spectral-embedding axis over genes, can turn the sparse, roughly 900-site-per-donor AHBA microarray data into continuous voxel-resolution expression maps for all 100 Alzheimer's risk genes simultaneously. The network learns a function $\Phi_\theta(x,y,z,c,\gamma(e_g))$ mapping MNI coordinates, a grey/white-matter label, and a positionally-encoded gene embedding to normalized expression; after training it can be evaluated at any voxel coordinate in the atlas. The spectral embedding axis — the first non-trivial eigenvector of the normalized Laplacian of the gene–gene correlation matrix — orders genes so that spatial neighbours in the axis have similar expression distributions, and this ordering is what lets a single network fit sparse data across many genes at once. The authors' evidence is indirect but consistent: averaging INR maps to the Desikan–Killiany atlas reproduces abagen's regional values (mean per-gene R=0.67, pooled R=0.70), and gene-wise correlations of the maps with ADNI tau-PET uptake track abagen's correlations at R=0.89, with established risk genes such as APOE, MAPT, ABCA1, and TOMM40 showing strong tau association. The paper explicitly acknowledges that the data violate the Nyquist–Shannon rate, so the unsampled voxels are filled by the network's implicit smoothness prior rather than by measurement.

Load-bearing premise

The load-bearing premise is that brain-wide gene expression is a smooth, continuous function of spatial coordinates, tissue class, and the gene embedding, even though the paper itself notes that the data violate the Nyquist–Shannon sampling rate, so every value at an unsampled voxel is produced by that smoothness assumption rather than by measurement.

Editorial extensions

If this is right

  • Any disease-relevant gene set can be mapped at voxel resolution by computing its own spectral-embedding axis, so the pipeline transfers to contexts beyond Alzheimer's disease.
  • Existing region-level imaging-transcriptomics workflows can be recomputed at higher resolution by averaging INR maps over any atlas, recovering abagen's regional values while retaining sub-regional detail.
  • Voxel-level maps enable analyses that regional aggregation hides, such as colocalizing gene expression with imaging biomarkers like tau-PET at the voxel scale.
  • The discordant genes the paper highlights (NEAT1, JAZF1, HSPH1, TMEM41A, PTK2B, ZNF184) show that the choice of interpolation can change a gene's apparent disease association, so reported gene–pathology links should state which interpolation method produced them.

Reading between the lines

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

  • A direct test of the smoothness prior would require point-wise expression measurements at unsampled locations, which the paper does not have; until such data exist, the regional validations cannot certify voxel-level accuracy.
  • The conditioning mechanism suggests a general recipe for under-sampled spatial signals: build an auxiliary axis that orders targets by similarity and let one network fit them jointly; the same trick could apply to PET tracers, cell-type densities, or other sparse brain maps.
  • A quantifiable consequence of the paper's regularization is that genes adjacent on the spectral-embedding axis should have more similar spatial maps than distant genes; this could be checked across all 100 gene pairs and against the axis distance, extending the paper's ANK3/SIRPA illustration.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes an Implicit Neural Representation (SIREN) approach, augmented by a one-dimensional spectral embedding of the gene correlation matrix, to interpolate sparse Allen Human Brain Atlas (AHBA) microarray samples into voxel-level, brain-wide expression maps for 100 Alzheimer's disease risk genes. The method is described in detail, including preprocessing via the abagen pipeline, a spectral gene axis, positional encoding, and a 12-layer SIREN architecture. The main empirical claims are (i) that the spectral embedding yields a meaningful ordinal gene axis, (ii) that INR voxel maps agree with abagen region-level interpolation (average per-gene R=0.67, pooled R=0.70), (iii) that these maps reproduce known associations between AD risk genes and tau-PET uptake in ADNI, and (iv) that the approach is robust to noise. The authors acknowledge in Sections 3.2 and 5.4 that the data violate the Nyquist-Shannon sampling rate and that no ground truth exists for voxel-level expression, relying instead on indirect assessments.

Significance. If the voxel-level maps were validated, the work would offer a useful tool for brain-wide spatial transcriptomics and for studying regional vulnerability in AD. The paper is transparent about its preprocessing and the free parameters of the model, and it provides extensive ablation studies. However, the central claim of quantitative voxel-level reliability is not established: both comparators (abagen and tau-PET) operate at the region level, and abagen is itself an interpolation of the same sparse samples, so agreement with abagen largely reflects shared smoothness assumptions. The paper contains no spatially held-out evaluation, no cross-donor validation, and no control for spatial autocorrelation in the tau analyses. The method may still be valuable as a regularized interpolation or hypothesis-generating tool, but the current evidence does not support the stronger 'super-resolution' claim. The limitations are honestly stated in Section 5.4, but they are load-bearing rather than peripheral.

major comments (5)
  1. [Section 4.5, Table 1, Figure 5] The primary quantitative validation compares INR predictions with abagen interpolations of the same sparse AHBA samples. The voxel-wise R>0.96 in Table 1 and the regional R=0.67-0.70 in Figure 5 are therefore measures of agreement between two interpolants, not accuracy at unsampled locations. Because both methods impose smoothness, correlated smoothness can inflate these correlations. The paper needs a spatially held-out evaluation: train on a subset of sample coordinates, predict the held-out coordinates, and report predictive error as a function of distance to the nearest training sample, comparing against abagen and a simple Gaussian-process or kriging baseline. Without such an experiment, the claim that the INR maps are quantitatively reliable at voxel level is unsupported.
  2. [Section 3.2, Section 5.4] The paper explicitly states in Section 3.2 that the data 'violates the Nyquist-Shannon sampling rate' and in Section 5.4 that 'voxel-level mapping of gene expression is difficult to validate, as there is no ground truth against which to compare our results.' These statements together imply that unsampled voxel values are supplied by the SIREN bandwidth regularization, not by data. The abstract and introduction nonetheless describe the output as 'fully resolved spatial maps' and 'super-resolution.' The manuscript should either reframe the contribution as regularized interpolation/conditioning, or provide independent validation against a dense spatial transcriptomics dataset (e.g., a slide or the BICCN atlas) where true voxel-level measurements exist for at least some genes.
  3. [Section 4.1, Section 4.4, Figure 2e, Figure S1] The spectral embedding axis is computed from the same gene expression matrix used to train the INR and to demonstrate that neighboring genes have similar spatial patterns. Sorting the correlation matrix by the first Laplacian eigenvector and then showing block structure is circular: the eigenvector is constructed to maximize such correlation. The claim that the gene axis is a 'natural disease axis' needs independent support. For example, the authors could show that the axis predicts gene ontology/pathway similarity, or that it orders genes consistently when computed on a held-out donor, or that it reproduces a known functional grouping not used in the embedding. As written, Section 4.4's ANK3/SIRPA example is constructed by the choice of embedding.
  4. [Section 4.6, Section 2.2, Figure 6] The tau-PET validation uses group-averaged ADNI tau over 86 Desikan-Killiany regions against expression from only two healthy donors, and the reported p-values treat these 86 regions as independent observations. This ignores strong spatial autocorrelation across neighboring regions, and no multiple-comparison correction is applied across the 100 genes tested. The weak correlations with Open Targets scores (R=0.12-0.14 in Figure 6c,d) are presented as a null result but are not discussed as a limitation of the tau-based validation. The authors should use spin-based permutation tests or other spatial-autocorrelation-preserving nulls, and report region-level effect sizes with confidence intervals rather than raw p-values.
  5. [Section 4.2, Figure 3] The train/validation/test split described in Section 4.2 is a random split over samples. Because gene expression is spatially autocorrelated, a random split assigns nearby samples to both training and test sets, making the reported low losses and high correlations partly a measure of spatial smoothness rather than generalization to unsampled locations. A spatial block cross-validation, where contiguous spatial regions are held out entirely, is necessary to estimate the model's true interpolation error. This is a load-bearing gap because the central claim concerns locations that were not sampled.
minor comments (6)
  1. [Section 5.1, paragraph 1] The word 'gens' should be 'genes' in the sentence 'at the other end are gens with a strong negative correlation with tau.'
  2. [Figure 6 caption] The caption contains the typo 'cooefficients' for 'coefficients.'
  3. [Figure 3 caption] The caption writes 'encodding dimension' and 'depther'; these should be 'encoding dimension' and 'depth.'
  4. [Supplementary Table 2] Several gene names contain spacing errors, e.g., 'NY AP1' instead of 'NYAP1' and 'nyeloid' instead of 'myeloid' in the TREM2 entry.
  5. [Section 3.1, Equation for spectral embedding] The notation for the eigenvector is inconsistent: the text uses v1 for both the eigenvector and the output value v; this should be disambiguated, for example by denoting the embedding as φ(g) as in Algorithm 1.
  6. [Data availability] The manuscript states that code will be shared 'upon paper acceptance'; providing the code and trained model weights at submission time would strengthen reproducibility and allow reviewers to test the interpolation claims.

Circularity Check

2 steps flagged · score 6.0 of 10

Gene-axis ordering is the spectral-embedding objective restated as a finding, and the abagen "external validation" compares two interpolants of the same sparse samples; the voxel-level super-resolution claim is therefore only partially independently supported.

  1. self definitional [Sections 3.1 and 4.1; Figs. 2e, S1b,d]
    "We performed spectral embedding of the gene correlation matrix (see Methods section 3) and obtained, using the first non-zero eigenvector of the matrix's Laplacian, a one-dimensional embedding of each gene in this spectral space. We sorted the genes in ascending order of this embedding. ... This reordering reveals a clearer structure in the heatmaps, with genes adjacent in the ordinal score axis exhibiting stronger spatial correlations with each other, as seen in the diagonally dominant pattern of the reordered matrices."

    The input feature eg is the first non-trivial eigenvector of the normalized Laplacian of a gene-affinity matrix built from the same expression values v that the INR is trained to reproduce. Sorting genes by this eigenvector is precisely the operation that makes correlated genes adjacent in the ordering; the reported 'clearer structure' and 'genes adjacent in the ordinal score axis exhibiting stronger spatial correlations' is the objective of the spectral embedding restated as an empirical result. The later claim that the gene axis regularizes the INR and aligns functionally related genes is therefore inherited from the construction of eg from the target data, not independently discovered by the model.

  2. fitted input called prediction [Section 4.5 (Numerical assessment of INR against abagen); Table 1; Fig. 5]
    "While cross-validation with the validation and test set are good indicators of the INR model's fit to samples, we also require external validation against a benchmark method. For this purpose, we again chose the widely-adopted abagen method as a comparison baseline since it provides regional interpolation based on gene probes, allowing us to assess if our interpolated gene values are reasonable."

    abagen's regional values are generated by inverse-distance interpolation of the same reannotated, noise-filtered AHBA probe data used to train the INR, as described in the Methods. After fitting the INR to those samples, the paper reports R=0.67 and R=0.70 agreement with abagen as evidence that the interpolated values are reasonable, but this is agreement between two smooth interpolants sharing the same input points, not a test against independent measurements at unsampled voxels.

full rationale

The INR regression itself is a self-contained supervised fit: it maps (x,y,z,c,eg) to v and is evaluated on held-out samples with reported losses, so the interpolation machinery is not inherently circular. Circularity enters at two points. First, the gene axis eg is the first non-trivial eigenvector of a Laplacian built from the gene-correlation matrix of the same expression values v; ordering genes by this eigenvector is exactly the operation that maximizes the banded/correlated structure later reported as a 'natural disease axis' and as evidence that neighboring genes have similar spatial patterns. That result is the defining objective of the embedding, not an independent discovery, and when fed back into the INR it functions as a compression of the target variable rather than an external constraint. Second, the 'external validation' against abagen compares two interpolants of the same AHBA samples, so the R=0.67/0.70 agreement is largely the expected overlap of smooth functions sharing their input points and does not test the unsampled voxel values that constitute the super-resolution claim. The paper explicitly concedes this validation gap in Section 5.4. There is no load-bearing self-citation chain or imported uniqueness theorem; the cited INR/SIREN machinery is external and the tau-PET analysis uses an independent dataset. The central construction of a learned smooth function on sparse samples is a legitimate method, but the paper's headline evidence for its reliability is partly constructed from the same data it is meant to verify.

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

The central product is a fitted deep network plus a one-dimensional gene embedding, both estimated from the same sparse AHBA matrix. There are no new physical entities, but there are four modeling choices that carry the result: the smoothness prior, the donor-representativeness assumption, the spectral embedding axis, and the tau group-averaging comparison. The absence of ground truth means these assumptions cannot be checked internally.

free parameters (5)
  • SIREN network weights and biases (12 layers, 512 units) = Trained on AHBA samples
    The central mapping is entirely determined by these fitted parameters; no independent estimate of voxel expression exists.
  • Positional encoding dimension N = 10
    Chosen from ablation experiments on the authors' data (Figure 3); affects the frequency bands the network can represent.
  • Learning rate, layer count, feature size, training steps = 1e-4, 12, 512, 5000
    Chosen via ablation on training loss; no theoretical justification is provided.
  • One-dimensional spectral embedding coordinate e_g = First non-trivial eigenvector of the normalized gene correlation Laplacian, normalized to [0,1]
    Computed from the same gene expression matrix used to train the model; the choice of a 1D projection and its use as a gene identity feature is a modeling decision.
  • Robust sigmoid normalization and min-max rescaling = Computed per sample and per gene
    Preprocessing choices affect the scale of all expression values and therefore all reported correlations.
assumptions (4)
  • domain assumption Gene expression varies smoothly as a continuous function of brain coordinates, grey or white matter label, and gene embedding.
    Invoked in Section 3.2 where INR is used as a low-bandwidth continuous representation; the paper admits the data violate Nyquist, so this prior generates all unsampled voxel values.
  • domain assumption AHBA samples from two donors are representative of healthy adult human brain baseline expression.
    Stated in Section 2.1 and acknowledged as a limitation in Section 5.4; all maps and tau correlations inherit donor-specific bias.
  • ad hoc to paper The gene spectral embedding from the expression correlation matrix provides a meaningful ordinal axis for spatial interpolation.
    Introduced in Section 3.1 and used to condition the INR; the axis is derived from the same data being interpolated rather than from independent biology.
  • domain assumption Group-averaged tau-PET uptake across ADNI subjects can be compared with healthy donor expression at the region level.
    Used in Section 4.6; there is no individual-level matching and no adjustment for age, sex, or spatial autocorrelation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Brain-wide interpolation and conditioning of gene expression in the human brain using Implicit Neural Representations." pith.science (2026). https://pith.science/paper/N56ARMAD

@misc{pith2026250611158,
  author       = {Pith},
  title        = {Pith review of: Brain-wide interpolation and conditioning of gene expression in the human brain using Implicit Neural Representations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N56ARMAD}},
  note         = {Machine review of arXiv:2506.11158}
}
read the original abstract

In this paper, we study the efficacy and utility of recent advances in non-local, non-linear image interpolation and extrapolation algorithms, specifically, ideas based on Implicit Neural Representations (INR), as a tool for analysis of spatial transcriptomics data. We seek to utilize the microarray gene expression data sparsely sampled in the healthy human brain, and produce fully resolved spatial maps of any given gene across the whole brain at a voxel-level resolution. To do so, we first obtained the 100 top AD risk genes, whose baseline spatial transcriptional profiles were obtained from the Allen Human Brain Atlas (AHBA). We adapted Implicit Neural Representation models so that the pipeline can produce robust voxel-resolution quantitative maps of all genes. We present a variety of experiments using interpolations obtained from Abagen as a baseline/reference.

Figures

Figures reproduced from arXiv: 2506.11158 by the authors.

Figure 1
Figure 1. Flow chart of the INR interpolation pipeline: Starting from the Allen Human Brain Atlas (left), which contains [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. a. Visualization of the spectral embedding pipeline: starting with raw gene expression data, calculating the affinity matrix representing gene-to-gene similarity, constructing the degree matrix, deriving the Laplacian matrix, and finally obtaining the normalized Laplacian used for embedding. b. Eigenvalue spectrum showing a clear spectral gap between the top eigenvalues and the rest. c. Scatter plot of gene position… view at source ↗
Figure 3
Figure 3. Comprehensive ablation study of model parameters and robustness analysis across two donors (9861 and 10021). [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: a: Sagittal view visualizations of gene expression patterns in the human brain using three different interpolation methods (INR + Spectral Embedding, INR + Spectral Embedding + Regional Averaged, and Abagen) for five tau-associated genes (ANK3, SIRPA, TSPOAP1, PRNP, an…
Figure 5
Figure 5. Figure 5: Region-wise comparison between INR and abagen. a: Individual genes correlation values between INR and abagen results ordered by spectral embedding, with an average Pearson correlation of 0.67 across all genes. b: Scatter plot comparing individual gene expression values…
Figure 6
Figure 6. Figure 6: Comparative tau correlation analysis of gene between [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

56 extracted references · 50 canonical work pages

  1. [1]

    Lein, E., Borm, L. E. & Linnarsson, S. The promise of spatial transcriptomics for neuroscience in the era of molecular cell typing.Sci. (New York, N.Y.)358, 64–69, DOI: 10.1126/science.aan6827 (2017). Place: United States

    work page doi:10.1126/science.aan6827 2017

  2. [2]

    Sci.10, 2206939, DOI: 10.1002/advs.202206939 (2023)

    Park, H.-E.et al.Spatial Transcriptomics: Technical Aspects of Recent Developments and Their Applications in Neuroscience and Cancer Research.Adv. Sci.10, 2206939, DOI: 10.1002/advs.202206939 (2023). Publisher: John Wiley & Sons, Ltd

    work page doi:10.1002/advs.202206939 2023

  3. [3]

    & Kim, T.-K

    Jung, N. & Kim, T.-K. Spatial transcriptomics in neuroscience.Exp. & Mol. Medicine55, 2105–2115, DOI: 10.1038/ s12276-023-01093-y (2023)

    work page 2023

  4. [4]

    & Meletis, K

    Ortiz, C., Carlén, M. & Meletis, K. Spatial Transcriptomics: Molecular Maps of the Mammalian Brain.Annu. review neuroscience44, 547–562, DOI: 10.1146/annurev-neuro-100520-082639 (2021). Place: United States

    work page doi:10.1146/annurev-neuro-100520-082639 2021

  5. [5]

    M.et al.Spatial transcriptomics in glioblastoma: is knowing the right zip code the key to the next therapeutic breakthrough?Front

    Shireman, J. M.et al.Spatial transcriptomics in glioblastoma: is knowing the right zip code the key to the next therapeutic breakthrough?Front. oncology13, 1266397, DOI: 10.3389/fonc.2023.1266397 (2023). Place: Switzerland

    arXiv 2023

  6. [6]

    Medicine24, 1910–1918, DOI: 10.1038/s41591-018-0206-4 (2018)

    Sepulcre, J.et al.Neurogenetic contributions to amyloid beta and tau spreading in the human cortex.Nat. Medicine24, 1910–1918, DOI: 10.1038/s41591-018-0206-4 (2018)

    work page doi:10.1038/s41591-018-0206-4 2018

  7. [7]

    Anand, C., Torok, J., Abdelnour, F., Maia, P. D. & Raj, A. Selective vulnerability and resilience to alzheimer’s disease tauopathy as a function of genes and the connectome.bioRxiv2024–03 (2024)

    work page 2024

  8. [8]

    Shafiei, G.et al.Network structure and transcriptomic vulnerability shape atrophy in frontotemporal dementia.Brain146, 321–336 (2023)

    work page 2023

Show all 56 references
  1. [9]

    D.et al.Conserved cell types with divergent features in human versus mouse cortex.Nature573, 61–68, DOI: 10.1038/s41586-019-1506-7 (2019)

    Hodge, R. D.et al.Conserved cell types with divergent features in human versus mouse cortex.Nature573, 61–68, DOI: 10.1038/s41586-019-1506-7 (2019)

    2019 doi

  2. [10]

    E.et al.Comparative cellular analysis of motor cortex in human, marmoset and mouse.Nature598, 111–119, DOI: 10.1038/s41586-021-03465-8 (2021)

    Bakken, T. E.et al.Comparative cellular analysis of motor cortex in human, marmoset and mouse.Nature598, 111–119, DOI: 10.1038/s41586-021-03465-8 (2021)

    2021 doi

  3. [11]

    & Duff, K

    Fu, H., Hardy, J. & Duff, K. E. Selective vulnerability in neurodegenerative diseases.Nat. neuroscience21, 1350–1358 (2018)

    2018

  4. [12]

    neuroscience24, 276–287 (2021)

    Leng, K.et al.Molecular characterization of selectively vulnerable neurons in alzheimer’s disease.Nat. neuroscience24, 276–287 (2021). 13.Gabitto, M. I.et al.Integrated multimodal cell atlas of alzheimer’s disease.Nat. Neurosci.27, 2366–2383 (2024)

    2021

  5. [14]

    neuroscience25, 588–595 (2022)

    Kamath, T.et al.Single-cell genomic profiling of human dopamine neurons identifies a population that selectively degenerates in parkinson’s disease.Nat. neuroscience25, 588–595 (2022)

    2022

  6. [15]

    J.et al.An anatomically comprehensive atlas of the adult human brain transcriptome.Nature489, 391–399, DOI: 10.1038/nature11405 (2012)

    Hawrylycz, M. J.et al.An anatomically comprehensive atlas of the adult human brain transcriptome.Nature489, 391–399, DOI: 10.1038/nature11405 (2012)

    2012 doi

  7. [16]

    J.et al.Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome.Proc

    Whitaker, K. J.et al.Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome.Proc. Natl. Acad. Sci. U. S. A.113, 9105–9110 (2016)

    2016

  8. [17]

    B.et al.Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography.Nat

    Burt, J. B.et al.Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography.Nat. Neurosci.21, 1251–1259, DOI: 10.1038/s41593-018-0195-0 (2018)

    2018 doi

  9. [18]

    D.et al.Standardizing workflows in imaging transcriptomics with the abagen toolbox.elife10, e72129 (2021)

    Markello, R. D.et al.Standardizing workflows in imaging transcriptomics with the abagen toolbox.elife10, e72129 (2021)

    2021

  10. [19]

    Gryglewski, G.et al.Spatial analysis and high resolution mapping of the human whole-brain transcriptome for integrative analysis in neuroimaging.NeuroImage176, DOI: 10.1016/j.neuroimage.2018.04.068 (2018)

    2018 doi

  11. [20]

    Compressed sensing.IEEE Transactions on Inf

    Donoho, D. Compressed sensing.IEEE Transactions on Inf. Theory52, 1289–1306, DOI: 10.1109/TIT.2006.871582 (2006). 21.Bora, A., Jalal, A., Price, E. & Dimakis, A. G. Compressed sensing using generative models (2017). 1703.03208. 18/28

    2006

  12. [22]

    D.et al.Standardizing workflows in imaging transcriptomics with the abagen toolbox.Elife10, DOI: 10.7554/eLife.72129 (2021)

    Markello, R. D.et al.Standardizing workflows in imaging transcriptomics with the abagen toolbox.Elife10, DOI: 10.7554/eLife.72129 (2021)

    2021 doi

  13. [23]

    M., Cruchaga, C

    Karch, C. M., Cruchaga, C. & Goate, A. M. Alzheimer’s disease genetics: from the bench to the clinic.Neuron83, 11–26, DOI: 10.1016/j.neuron.2014.05.041 (2014)

    2014 doi

  14. [24]

    A., Wagner, C

    Ayoub, C. A., Wagner, C. S. & Kuret, J. Identification of gene networks mediating regional resistance to tauopathy in late-onset alzheimer’s disease.PLoS Genet.19, e1010681, DOI: 10.1371/journal.pgen.1010681 (2023)

    2023 doi

  15. [25]

    L., Dumitrescu, L

    Seto, M., Weiner, R. L., Dumitrescu, L. & Hohman, T. Protective genes and pathways in alzheimer’s disease: moving towards precision interventions.Mol Neurodegener16, 16, DOI: https://doi.org/10.1186/s13024-021-00452-5 (2021)

    2021 doi

  16. [26]

    J., Fulton-Howard, B

    Andrews, S. J., Fulton-Howard, B. & Goate, A. Interpretation of risk loci from genome-wide association studies of alzheimer’s disease.Lancet Neurol19, 326–335, DOI: 10.1016/S1474-4422(19)30435-1 (2020)

    2020 doi

  17. [27]

    Dumitrescu, L.et al.Genetic variants and functional pathways associated with resilience to alzheimer’s disease.Brain 143, 2561–2575, DOI: 10.1093/brain/awaa209 (2020)

    2020 doi

  18. [28]

    Arnatkeviciute, A., Fulcher, B. D. & Fornito, A. A practical guide to linking brain-wide gene expression and neuroimaging data.Neuroimage189, 353–367, DOI: 10.1016/j.neuroimage.2019.01.011 (2019)

    2019 doi

  19. [29]

    30.Hawrylycz, M.et al.Canonical genetic signatures of the adult human brain.Nat

    Pandya, S.et al.Modeling seeding and neuroanatomic spread of pathology in amyotrophic lateral sclerosis.Neuroimage 251, 118968, DOI: 10.1016/j.neuroimage.2022.118968 (2022). 30.Hawrylycz, M.et al.Canonical genetic signatures of the adult human brain.Nat. neuroscience18, 1832–1...

    2022

  20. [31]

    Romero-Garcia, R.et al.Structural covariance networks are coupled to expression of genes enriched in supragranular layers of the human cortex.Neuroimage171, 256–267 (2018)

    2018

  21. [32]

    Desikan, R. S.et al.An automated labeling system for subdividing the human cerebral cortex on mri scans into gyral based regions of interest.Neuroimage31, 968–80, DOI: 10.1016/j.neuroimage.2006.01.021 (2006). 33.Pedregosa, F.et al.Scikit-learn: Machine learning in Python.J. Ma...

    2006 doi

  22. [34]

    & Weiss, Y

    Ng, A., Jordan, M. & Weiss, Y . On spectral clustering: Analysis and an algorithm. In Dietterich, T., Becker, S. & Ghahramani, Z. (eds.)Advances in Neural Information Processing Systems, vol. 14 (MIT Press, 2001)

    2001

  23. [35]

    Sitzmann, V ., Martel, J. N. P., Bergman, A. W., Lindell, D. B. & Wetzstein, G. Implicit neural representations with periodic activation functions (2020). 2006.09661. 36.Mildenhall, B.et al.Nerf: Representing scenes as neural radiance fields for view synthesis (2020). 2003.089...

    2020 arXiv

  24. [38]

    PMID: 27899665; PMCID: PMC5210543

    Koscielny, G.et al.Open targets: a platform for therapeutic target identification and validation.Nucleic Acids Res.45, D985–D994, DOI: 10.1093/nar/gkw1055 (2017). PMID: 27899665; PMCID: PMC5210543

    2017 doi

  25. [39]

    Janson, J.et al.Increased risk of type 2 diabetes in alzheimer disease.Diabetes53, 474–81, DOI: 10.2337/diabetes.53.2.474 (2004)

    2004 doi

  26. [40]

    & Kötter, R

    Sporns, O., Tononi, G. & Kötter, R. The human connectome: a structural description of the human brain.PLoS computational biology1, e42 (2005). 41.Oh, S. W.et al.A mesoscale connectome of the mouse brain.Nature508, 207–214 (2014)

    2005

  27. [42]

    Bullmore, E. T. & Bassett, D. S. Brain graphs: graphical models of the human brain connectome.Annu. review clinical psychology7, 113–140 (2011). 43.Zeng, H. Mesoscale connectomics.Curr. opinion neurobiology50, 154–162 (2018)

    2011

  28. [44]

    French, L., Tan, P. P. C. & Pavlidis, P. Large-scale analysis of gene expression and connectivity in the rodent brain: insights through data integration.Front. neuroinformatics5, 12 (2011)

    2011

  29. [45]

    & Deng, H

    Ji, S., Fakhry, A. & Deng, H. Integrative analysis of the connectivity and gene expression atlases in the mouse brain. NeuroImage84, 245–253 (2014)

    2014

  30. [46]

    Tan, P. P. C., French, L. & Pavlidis, P. Neuron-enriched gene expression patterns are regionally anti-correlated with oligodendrocyte-enriched patterns in the adult mouse and human brain.Front. neuroscience7, 5 (2013)

    2013

  31. [47]

    & Wronkiewicz, M

    Henriksen, S., Pang, R. & Wronkiewicz, M. A simple generative model of the mouse mesoscale connectome.Elife5, e12366 (2016). 19/28

    2016

  32. [48]

    Fulcher, B. D. & Fornito, A. A transcriptional signature of hub connectivity in the mouse connectome.Proc. Natl. Acad. Sci.113, 1435–1440 (2016)

    2016

  33. [49]

    W.et al.A null model of the mouse whole-neocortex micro-connectome.Nat

    Reimann, M. W.et al.A null model of the mouse whole-neocortex micro-connectome.Nat. communications10, 3903 (2019)

    2019

  34. [50]

    & Raj, A

    Goel, P., Kuceyeski, A., LoCastro, E. & Raj, A. Spatial patterns of genome-wide expression profiles reflect anatomic and fiber connectivity architecture of healthy human brain.Hum. brain mapping35, 4204–4218 (2014)

    2014

  35. [51]

    E.et al.Gene transcription profiles associated with inter-modular hubs and connection distance in human functional magnetic resonance imaging networks.Philos

    Vértes, P. E.et al.Gene transcription profiles associated with inter-modular hubs and connection distance in human functional magnetic resonance imaging networks.Philos. Transactions Royal Soc. B: Biol. Sci.371, 20150362 (2016)

    2016

  36. [52]

    & Sepulcre, J

    Diez, I. & Sepulcre, J. Neurogenetic profiles delineate large-scale connectivity dynamics of the human brain.Nat. communications9, 3876 (2018)

    2018

  37. [53]

    D., Markley, E

    Mezias, C., Torok, J., Maia, P. D., Markley, E. & Raj, A. Matrix inversion and subset selection (miss): A pipeline for mapping of diverse cell types across the murine brain.Proc. Natl. Acad. Sci.119, e2111786119 (2022)

    2022

  38. [54]

    Sun, Y .et al.A transcriptome-wide association study of alzheimer’s disease using prediction models of relevant tissues identifies novel candidate susceptibility genes.Genome medicine13, 1–11 (2021)

    2021

  39. [55]

    Seshadri, S.et al.Genome-wide analysis of genetic loci associated with alzheimer disease.Jama303, 1832–1840 (2010)

    2010

  40. [56]

    genetics54, 412–436 (2022)

    Bellenguez, C.et al.New insights into the genetic etiology of alzheimer’s disease and related dementias.Nat. genetics54, 412–436 (2022)

    2022

  41. [57]

    W.et al.Genetic meta-analysis of diagnosed alzheimer’s disease identifies new risk loci and implicates a β, tau, immunity and lipid processing.Nat

    Kunkle, B. W.et al.Genetic meta-analysis of diagnosed alzheimer’s disease identifies new risk loci and implicates a β, tau, immunity and lipid processing.Nat. genetics51, 414–430 (2019)

    2019

  42. [58]

    Fusco, F. R.et al.Cellular localization of huntingtin in striatal and cortical neurons in rats: lack of correlation with neuronal vulnerability in huntington’s disease.J Neurosci19, 1189–202, DOI: 10.1523/JNEUROSCI.19-04-01189.1999 (1999)

    1999

  43. [59]

    Selective neuronal death in neurodegenerative diseases: The ongoing mystery.Yale J Biol Med92, 695–705 (2019)

    Subramaniam, S. Selective neuronal death in neurodegenerative diseases: The ongoing mystery.Yale J Biol Med92, 695–705 (2019)

    2019

  44. [60]

    & Wang, X

    Chen, Y ., Liu, S. & Wang, X. Learning continuous image representation with local implicit image function. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 8628–8638 (2021)

    2021

  45. [61]

    InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 20991–21002 (2023)

    Sun, J.et al.Next3d: Generative neural texture rasterization for 3d-aware head avatars. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 20991–21002 (2023)

    2023

  46. [62]

    & Kanazawa, A

    Yu, A., Ye, V ., Tancik, M. & Kanazawa, A. pixelnerf: Neural radiance fields from one or few images. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 4578–4587 (2021)

    2021

  47. [63]

    & Geiger, A

    Niemeyer, M., Mescheder, L., Oechsle, M. & Geiger, A. Differentiable volumetric rendering: Learning implicit 3d representations without 3d supervision. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 3504–3515 (2020)

    2020

  48. [64]

    & Singh, V

    Pal, S., Adepu, H., Wang, C., Golland, P. & Singh, V . Implicit representations via operator learning. InForty-first International Conference on Machine Learning(2024). 65.Siletti, K.et al.Transcriptomic diversity of cell types across the adult human brain.Science382, eadd7046...

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.