pith. sign in

arxiv: 2606.05870 · v1 · pith:VWRXILVAnew · submitted 2026-06-04 · 🧬 q-bio.NC · cs.LG· q-bio.QM

Cross-scale spatially-aware generative modeling of transcriptomic programs underlying neurodegenerative brain organization

Krishnakumar Vaithianathan (for the Alzheimer's Disease Neuroimaging Initiative) This is my paper

Pith reviewed 2026-06-27 22:59 UTC · model grok-4.3

classification 🧬 q-bio.NC cs.LGq-bio.QM
keywords neurodegenerationtranscriptomicsgenerative modelingAlzheimer's diseasecortical thinningvariational architecturespatial regularizationimaging transcriptomics
0
0 comments X

The pith

A variational generative model trained on regional gene expression predicts Alzheimer's cortical thinning with 0.86 explained variance.

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

The paper introduces a cross-scale generative framework that learns latent transcriptomic programs from gene expression profiles across 68 cortical regions to model how molecular organization produces patterns of cortical degeneration. Regional transcriptomic data from the Allen Atlas using 910 landmark genes is combined with ADNI-derived maps of cortical thinning between controls and Alzheimer's subjects. A variational architecture with graph-based spatial smoothness regularization is trained to link these scales, achieving an explained variance of 0.8604 and spatial correlation r=0.9439. If correct, this shows that biologically constrained generative models can extract structured representations of disease susceptibility rather than relying on direct correlations between genes and imaging phenotypes.

Core claim

The central claim is that a spatially-aware variational generative model can learn latent biological programs from regional transcriptomic profiles that predict regional neurodegenerative vulnerability, demonstrated by strong prediction performance on ADNI cortical thinning data and structured latent representations tied to distributed disease susceptibility.

What carries the argument

Variational generative architecture with graph-based spatial smoothness regularization, trained on 68-region transcriptomic profiles to predict cortical degeneration maps.

If this is right

  • The learned latent representations identify structured transcriptomic organization linked to distributed disease susceptibility.
  • Biologically constrained generative modeling can bridge microscale molecular data with macroscale neurodegeneration patterns.
  • This approach supplies a foundation for spatially-aware generative neurobiology beyond correlation-based imaging-transcriptomic studies.
  • Regional vulnerability maps can be generated from transcriptomic input while preserving cortical spatial organization.

Where Pith is reading between the lines

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

  • The same framework could be retrained on data from other neurodegenerative diseases to test whether similar latent programs emerge for different vulnerability patterns.
  • If the latent programs prove stable, they could be used to prioritize genes for functional validation in cellular models of neurodegeneration.
  • Extending the model to include longitudinal imaging might allow prediction of individual progression rates from baseline transcriptomic profiles.
  • The spatial regularization term could be adapted to incorporate white-matter connectivity to test whether network topology further improves predictions.

Load-bearing premise

The model extracts biologically meaningful latent programs that explain cortical thinning patterns rather than fitting noise or dataset-specific artifacts in the ADNI and Allen Atlas data.

What would settle it

Testing the trained model on an independent cohort with matched gene expression and cortical thickness data and finding substantially lower explained variance or spatial correlation than 0.86 and 0.94.

Figures

Figures reproduced from arXiv: 2606.05870 by Krishnakumar Vaithianathan (for the Alzheimer's Disease Neuroimaging Initiative).

Figure 1
Figure 1. Figure 1: Overview of the transcriptomic preprocessing workflow used in the present study. Regional gene expression [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Spatial organization of Alzheimer’s disease-related cortical neurodegeneration vulnerability across 68 cortical [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Transcriptomic organization of landmark genes across cortical regions. Variance analyses revealed substantial [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Low-dimensional visualization of latent gene program organization learned by the spatially-aware generative [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Prediction performance of the proposed cross-scale generative framework. The model demonstrated strong [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
read the original abstract

Neurodegenerative disorders such as Alzheimer's disease exhibit highly organized patterns of regional brain vulnerability, yet the biological mechanisms underlying this spatial selectivity remain incompletely understood. Existing imaging-transcriptomic studies have largely relied on correlation-based analyses between gene expression and neuroimaging phenotypes, limiting their ability to model how molecular organization gives rise to neurodegeneration. Here, we introduce a cross-scale spatially-aware generative framework for modeling transcriptomic programs underlying cortical neurodegeneration. Regional transcriptomic profiles were derived from the Allen Human Brain Atlas using 910 landmark genes across 68 cortical regions. Neurodegenerative vulnerability maps were constructed from ADNI FreeSurfer cortical thickness measurements by computing regional cortical thinning differences between cognitively normal controls (NC = 926) and Alzheimer's disease subjects (AD = 426). A variational generative architecture was used to learn latent biological programs linking regional gene-expression organization to cortical degeneration while incorporating graph-based spatial smoothness regularization to preserve cortical organization. The proposed framework achieved strong prediction of regional neurodegenerative vulnerability, yielding an explained variance of 0.8604 and a significant spatial correlation between predicted and observed cortical degeneration profiles (r = 0.9439, p < 0.001). The learned latent representations revealed structured transcriptomic organization associated with distributed disease susceptibility. These findings demonstrate that biologically constrained generative modeling can bridge microscale molecular organization with macroscale neurodegeneration, providing a foundation for spatially-aware generative neurobiology and computational neuroscience.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript introduces a cross-scale spatially-aware variational generative model that maps regional transcriptomic profiles (910 landmark genes across 68 cortical regions from the Allen Human Brain Atlas) to Alzheimer's disease-related cortical thinning vulnerability maps derived from ADNI FreeSurfer data (NC=926, AD=426). The model incorporates graph-based spatial smoothness regularization to learn latent biological programs, and reports strong predictive performance with explained variance of 0.8604 and spatial correlation r=0.9439 (p<0.001) between predicted and observed degeneration profiles, plus analysis of structured latent representations linked to disease susceptibility.

Significance. If the performance metrics reflect genuine out-of-sample predictive power rather than in-sample fitting, the framework would represent a meaningful advance over correlation-based imaging-transcriptomic analyses by using generative modeling to bridge microscale gene expression with macroscale neurodegeneration patterns while respecting cortical topology via spatial regularization.

major comments (1)
  1. [Abstract] Abstract: The reported explained variance (0.8604) and spatial correlation (r=0.9439) are presented as results of 'strong prediction' of regional neurodegenerative vulnerability, yet the abstract supplies no information on validation strategy (train/test split, k-fold cross-validation, permutation baseline, or external cohort). With only 68 regions as samples and a variational generative architecture, these metrics are load-bearing for the central claim but cannot be interpreted as evidence of generalizable prediction without such details.
minor comments (1)
  1. [Abstract] Abstract: The description of the generative architecture and graph regularization is high-level; a brief statement of the latent dimension or loss components would improve clarity for readers.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the abstract. We address the point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The reported explained variance (0.8604) and spatial correlation (r=0.9439) are presented as results of 'strong prediction' of regional neurodegenerative vulnerability, yet the abstract supplies no information on validation strategy (train/test split, k-fold cross-validation, permutation baseline, or external cohort). With only 68 regions as samples and a variational generative architecture, these metrics are load-bearing for the central claim but cannot be interpreted as evidence of generalizable prediction without such details.

    Authors: We agree that the abstract should briefly indicate the validation strategy to support interpretation of the performance metrics. The Methods section of the manuscript describes the evaluation procedure (including cross-validation and significance testing), but this information is not summarized in the abstract. We will revise the abstract to include a concise statement on the validation approach used. With respect to the sample size of 68 regions, this corresponds to the complete cortical parcellation; the graph-based spatial regularization is intended to improve robustness by incorporating topological constraints, though we acknowledge that external validation on independent cohorts would further strengthen generalizability claims. revision: yes

Circularity Check

1 steps flagged

In-sample fit on 68 regions presented as 'prediction' of vulnerability

specific steps
  1. fitted input called prediction [Abstract]
    "The proposed framework achieved strong prediction of regional neurodegenerative vulnerability, yielding an explained variance of 0.8604 and a significant spatial correlation between predicted and observed cortical degeneration profiles (r = 0.9439, p < 0.001)."

    Regional transcriptomic profiles from 68 cortical areas and the vulnerability maps derived from the same ADNI cohort are the sole inputs; the variational model is trained to link them, so the quoted 'prediction' metrics are the training-set reconstruction statistics rather than out-of-sample results.

full rationale

The paper's central performance claim rests on calling the output of a variational model trained to map 910-gene profiles from 68 regions onto the corresponding ADNI-derived thinning map a 'prediction' that yields R²=0.8604 and r=0.9439. Because the abstract and described data construction use exactly the same 68 regions for both the generative training and the reported metrics, with no indication of a held-out set, the numbers reduce to in-sample fit statistics by construction. This matches the fitted_input_called_prediction pattern; the remainder of the derivation (latent programs, graph regularization) does not alter that reduction for the headline metrics.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no equations, methods section, or implementation details, so free parameters, axioms, and invented entities cannot be enumerated; the model description implies standard variational assumptions and a domain assumption of spatial smoothness but supplies no explicit list or justification.

pith-pipeline@v0.9.1-grok · 5793 in / 1501 out tokens · 41084 ms · 2026-06-27T22:59:30.496452+00:00 · methodology

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

55 extracted references · 1 canonical work pages

  1. [1]

    Acta Neuropathologica , volume=

    Neuropathological stageing of Alzheimer-related changes , author=. Acta Neuropathologica , volume=. 1991 , doi=

    1991

  2. [2]

    Neuron , volume=

    Neurodegenerative diseases target large-scale human brain networks , author=. Neuron , volume=. 2009 , doi=

    2009

  3. [3]

    Alzheimer's & Dementia , volume=

    NIA-AA Research Framework: Toward a biological definition of Alzheimer's disease , author=. Alzheimer's & Dementia , volume=. 2018 , doi=

    2018

  4. [4]

    and Bernstein, MA

    Jack Jr., CR. and Bernstein, MA. and Fox, NC. and Thompson, P. and Alexander, G. and Harvey, D. and Borowski, B. and Britson, PJ. and L. Whitwell, J. and Ward, C. and Dale, AM. and Felmlee, JP. and Gunter, JL. and Hill, DLG. and Killiany, R. and Schuff, N. and Fox-Bosetti, S. and Lin, C. and Studholme, C. and DeCarli, CS. and Gunnar Krueger. and Ward, HA....

    2008

  5. [5]

    Journal of Neuroscience , volume=

    Dynamics of gray matter loss in Alzheimer's disease , author=. Journal of Neuroscience , volume=. 2003 , doi=

    2003

  6. [6]

    Neuron , volume=

    A network diffusion model of disease progression in dementia , author=. Neuron , volume=. 2012 , doi=

    2012

  7. [7]

    Cell reports , volume=

    Network diffusion model of progression predicts longitudinal patterns of atrophy and metabolism in Alzheimer’s disease , author=. Cell reports , volume=. 2015 , publisher=

    2015

  8. [8]

    Neuron , volume=

    Predicting regional neurodegeneration from the healthy brain functional connectome , author=. Neuron , volume=. 2012 , doi=

    2012

  9. [9]

    Nature Communications , volume=

    Spread of pathological tau proteins through communicating neurons in human Alzheimer's disease , author=. Nature Communications , volume=. 2020 , doi=

    2020

  10. [10]

    Nature , volume=

    An anatomically comprehensive atlas of the adult human brain transcriptome , author=. Nature , volume=. 2012 , doi=

    2012

  11. [11]

    Trends in Cognitive Sciences , volume=

    Bridging the gap between connectome and transcriptome , author=. Trends in Cognitive Sciences , volume=. 2019 , doi=

    2019

  12. [12]

    Journal of Neuroimaging , volume=

    Imaging transcriptomics in neurodegenerative diseases , author=. Journal of Neuroimaging , volume=. 2021 , publisher=

    2021

  13. [13]

    NeuroImage , volume=

    A systematic review of imaging transcriptomics studies in the human brain , author=. NeuroImage , volume=. 2021 , doi=

    2021

  14. [14]

    Biological Psychiatry , volume=

    Toward best practices for imaging transcriptomics of the human brain , author=. Biological Psychiatry , volume=. 2023 , publisher=

    2023

  15. [15]

    Elife , volume=

    Integrated transcriptomic and neuroimaging brain model decodes biological mechanisms in aging and Alzheimer’s disease , author=. Elife , volume=. 2021 , publisher=

    2021

  16. [16]

    Brain , volume=

    Imaging cortical propagation of neurodegeneration in Alzheimer's disease , author=. Brain , volume=. 2018 , doi=

    2018

  17. [17]

    eLife , volume=

    Standardizing workflows in imaging transcriptomics with the abagen toolbox , author=. eLife , volume=. 2021 , doi=

    2021

  18. [18]

    NeuroImage , volume=

    FreeSurfer , author=. NeuroImage , volume=. 2012 , doi=

    2012

  19. [19]

    International Conference on Learning Representations , year=

    Auto-Encoding Variational Bayes , author=. International Conference on Learning Representations , year=

  20. [20]

    arXiv preprint arXiv:1606.05908 , year=

    Tutorial on Variational Autoencoders , author=. arXiv preprint arXiv:1606.05908 , year=

    work page arXiv

  21. [21]

    MICCAI workshop on deep generative models , pages=

    Brain imaging generation with latent diffusion models , author=. MICCAI workshop on deep generative models , pages=. 2022 , organization=

    2022

  22. [22]

    Frontiers in artificial intelligence , volume=

    Generative models of brain dynamics , author=. Frontiers in artificial intelligence , volume=. 2022 , publisher=

    2022

  23. [23]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Deeper insights into graph convolutional networks for semi-supervised learning , author=. Proceedings of the AAAI conference on artificial intelligence , volume=. 2018 , doi=

    2018

  24. [24]

    IEEE Access , volume=

    A comprehensive survey on geometric deep learning , author=. IEEE Access , volume=. 2020 , publisher=

    2020

  25. [25]

    Medical image analysis , volume=

    Disease prediction using graph convolutional networks: application to autism spectrum disorder and Alzheimer’s disease , author=. Medical image analysis , volume=. 2018 , publisher=

    2018

  26. [26]

    Trends in Cognitive Sciences , volume=

    Large-scale gradients in human cortical organization , author=. Trends in Cognitive Sciences , volume=. 2018 , doi=

    2018

  27. [27]

    PNAS , volume=

    Situating the default-mode network along a principal gradient of macroscale cortical organization , author=. PNAS , volume=. 2016 , doi=

    2016

  28. [28]

    NeuroImage , volume=

    An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest , author=. NeuroImage , volume=. 2006 , doi=

    2006

  29. [29]

    Cell , volume=

    A Next Generation Connectivity Map: L1000 platform and the first 1,000,000 profiles , author=. Cell , volume=. 2017 , doi=

    2017

  30. [30]

    International Journal of Molecular Sciences , volume=

    Morphometric Similarity Patterning of Amyloid-β and Tau Proteins Correlates with Transcriptomics in the Alzheimer's Disease Continuum , author=. International Journal of Molecular Sciences , volume=. 2024 , doi=

    2024

  31. [31]

    Molecular Neurobiology , year=

    A bibliometric analysis of the spatial transcriptomics literature from 2006 to 2023 , author=. Molecular Neurobiology , year=

    2006

  32. [32]

    Neuron , volume=

    Accelerating biomedical discoveries in brain health through transformative neuropathology of aging and neurodegeneration , author=. Neuron , volume=. 2025 , publisher=

    2025

  33. [33]

    Frontiers in Big Data , volume=

    A review of AI-based radiogenomics in neurodegenerative disease , author=. Frontiers in Big Data , volume=. 2025 , doi=

    2025

  34. [34]

    Frontiers in Aging Neuroscience , volume=

    Advancements in multi-omics research to address challenges in Alzheimer's disease , author=. Frontiers in Aging Neuroscience , volume=. 2025 , doi=

    2025

  35. [35]

    bioRxiv , year=

    Single-cell digital twins identify drug targets and repurposable medicine in Alzheimer's disease , author=. bioRxiv , year=

  36. [36]

    bioRxiv , year=

    Cell type transcriptomics reveal shared genetic mechanisms in Alzheimer's and Parkinson's disease , author=. bioRxiv , year=

  37. [37]

    bioRxiv , pages=

    BrainBeacon: A Cross-Species Foundation Model for Single-cell Resolved Brain Spatial Transcriptomics , author=. bioRxiv , pages=. 2025 , publisher=

    2025

  38. [38]

    bioRxiv , pages=

    Tissue reassembly with generative AI , author=. bioRxiv , pages=. 2025 , publisher=

    2025

  39. [39]

    Frontiers in Dementia , volume=

    Neurogenesis drives hippocampal formation-wide spatial transcription alterations in health and Alzheimer's disease , author=. Frontiers in Dementia , volume=. 2025 , doi=

    2025

  40. [40]

    Journal of Neural Engineering , volume=

    An interpretable generative multimodal neuroimaging-genomics framework for decoding Alzheimer’s disease , author=. Journal of Neural Engineering , volume=. 2025 , publisher=

    2025

  41. [41]

    Nature , volume=

    Deep learning , author=. Nature , volume=. 2015 , doi=

    2015

  42. [42]

    NeurIPS , volume=

    Generative adversarial nets , author=. NeurIPS , volume=. 2014 , doi=

    2014

  43. [43]

    NeurIPS , year=

    Attention Is All You Need , author=. NeurIPS , year=

  44. [44]

    Nature methods , pages=

    Nicheformer: a foundation model for single-cell and spatial omics , author=. Nature methods , pages=. 2025 , publisher=

    2025

  45. [45]

    Nature , volume=

    Alpha-synuclein in Lewy bodies , author=. Nature , volume=. 1997 , doi=

    1997

  46. [46]

    Neuron , volume=

    Neurotransmitter release: the last millisecond in the life of a synaptic vesicle , author=. Neuron , volume=. 2013 , doi=

    2013

  47. [47]

    Science , volume=

    Activity-dependent regulation of MEF2 transcription factors suppresses excitatory synapse number , author=. Science , volume=. 2006 , doi=

    2006

  48. [48]

    Pharmacological Research , volume=

    The characterization of microtubule-stabilizing drugs as possible therapeutic agents for Alzheimer's disease and related tauopathies , author=. Pharmacological Research , volume=. 2011 , doi=

    2011

  49. [49]

    eLife , volume=

    MEF2C regulates cortical inhibitory and excitatory synapses and behaviors relevant to neurodevelopmental disorders , author=. eLife , volume=. 2018 , doi=

    2018

  50. [50]

    , author=

    Visualizing data using t-SNE. , author=. Journal of machine learning research , volume=

  51. [51]

    Nature Neuroscience , volume=

    Network neuroscience , author=. Nature Neuroscience , volume=. 2017 , doi=

    2017

  52. [52]

    Nature Communications , volume=

    Transcriptomic and cellular decoding of regional brain vulnerability to neurogenetic disorders , author=. Nature Communications , volume=. 2020 , doi=

    2020

  53. [53]

    NeuroImage , volume=

    Partial Least Squares (PLS) methods for neuroimaging: a tutorial and review , author=. NeuroImage , volume=. 2011 , doi=

    2011

  54. [54]

    IEEE Signal Processing Magazine , volume=

    Geometric deep learning: going beyond Euclidean data , author=. IEEE Signal Processing Magazine , volume=. 2017 , doi=

    2017

  55. [55]

    Advances in Neural Information Processing Systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in Neural Information Processing Systems , volume=. 2020 , doi=

    2020