Pith. sign in

REVIEW 4 major objections 4 minor 43 references

DuAL-Net: A Hybrid Framework for Alzheimer's Disease Prediction from Whole-Genome Sequencing via Local SNP Windows and Global Annotations

T0 review · 4 major / 4 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 DuAL-Net predicts Alzheimer's dementia from whole-genome sequencing by fusing local SNP-window probabilities with global functional annotations, reporting an average AUC of 0.671 for top-ranked SNP panels, 20.3 percent above randomly…

desk verdict Useful hybrid architecture, but the central ranking claim is undermined by selection bias: the SNP ranking and the AUC evaluation share the same 1,050 samples. read the letter →

arxiv 2506.00673 v1 pith:5BJSX4Y2 submitted 2025-05-31 q-bio.GN

Eun Hye Lee , Taeho Jo This is my paper
classification q-bio.GN
keywords Alzheimer'sdiseasewhole-genomesequencingSNPprioritizationTabNetRandomForeststackingensembleAPOEregiongenomicprediction
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

DuAL-Net is a two-branch deep-learning framework for predicting Alzheimer's dementia from whole-genome sequencing. The local branch scores short SNP windows with an out-of-fold stacking ensemble of TabNet and Random Forest; the global branch scores SNPs by the predictive accuracy of annotation-defined groups, and the two scores are blended with a weighting parameter. The paper's central claim is that the resulting SNP ranking carries real predictive signal: top-ranked panels of 100, 500, and 1,000 SNPs reach an average cross-validated AUC of 0.671, beating randomly selected panels (0.558) and bottom-ranked panels (0.497), and the top SNPs include well-established APOE variants. A sympathetic reader would care because a validated ranking would turn WGS into an interpretable, early risk-screening tool. The paper itself notes that the analysis is confined to a one-megabase APOE-centered region, that the sample size is modest, and that annotation resources may bias attention toward known loci.

What carries the argument

The load-bearing machinery is DuAL-Net's dual scoring pipeline. The local branch breaks the APOE-region WGS data into non-overlapping windows, trains a TabNet model (a deep network with sequential attention for tabular data) and a Random Forest on each window, and combines their out-of-fold predictions with a logistic regression meta-model. The global branch converts SNP annotations into binary indicators, trains the same TabNet/Random Forest stack on each annotation-defined group, and assigns each SNP the average accuracy of all groups it belongs to. A single parameter alpha (optimized to 0.8 by cross-validated AUC) blends the local and global scores, producing the ranking used to select SNP panels.

What would settle it

Re-run DuAL-Net with SNP ranking determined only inside each training fold (or on a separate discovery cohort) and evaluate the top-ranked panels solely on held-out samples; if the top panels no longer beat randomly selected panels, the claimed prioritization signal is an artifact of using test labels during feature selection.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that combining local and global genomic signals improves AD prediction over either alone. With alpha=0.8, the integrated model reaches 0.678 cross-validated accuracy, versus 0.605 for the global-only model and 0.677 for the local-only model. Ranking all 14,011 SNPs by the combined score and taking the top 100 yields AUC 0.697; the average AUC across 100, 500, and 1,000 SNP panels is 0.671, compared with 0.497 for bottom-ranked and 0.558 for randomly selected SNPs. The top-100 SNPs fall in APOE and APOC1, and the top-500 set adds TOMM40 and NECTIN2, including known AD-risk variants such as rs429358 and rs7412.

Load-bearing premise

The central claim assumes that ranking SNPs from the full dataset and then cross-validating AUC on the same dataset does not leak label information into the selection step; if it does, the reported advantage of top-ranked over randomly selected SNPs would not generalize to new data.

Editorial extensions

If this is right

  • If the ranking is valid, DuAL-Net gives researchers a supervised SNP-prioritization tool that needs no external outcome labels beyond the training cohort.
  • The recovered top SNPs matching established APOE, TOMM40, and NECTIN2 associations would support the biological relevance of the prioritization and could focus follow-up genotyping.
  • Because the framework's two branches can incorporate additional annotation categories or external modalities, the same stacking design could transfer to other complex diseases with WGS data.
  • The reported gap between top-ranked and random panels would give a quantitative benchmark for future WGS-based AD risk models.

Reading between the lines

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

  • The paper does not test whether the SNP ranking would survive nested cross-validation or an independent cohort; moving the ranking step inside each training fold would be the direct test of whether the top-versus-random AUC gap generalizes.
  • Because all SNPs come from a single APOE-centered megabase, the 'global' annotation branch is global in feature type, not in genome coverage; whether the long-range modeling claim holds genome-wide is an open extension.
  • The alpha=0.8 optimum suggests the local window signal dominates, but the global branch still adds a small accuracy gain; a learned or data-dependent alpha could replace the grid search on larger datasets.
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

4 major / 4 minor

Summary. The manuscript proposes DuAL-Net, a hybrid framework for Alzheimer's disease (AD) prediction from whole-genome sequencing data. The method segments the genome into non-overlapping local windows and separately groups SNPs by functional annotations, using out-of-fold stacking of TabNet and Random Forest classifiers to produce local and global accuracy scores that are combined with a weighting parameter α. The authors evaluate the approach on 1,050 individuals (443 cognitively normal, 607 AD dementia) using five-fold cross-validation on a 1-Mb region around APOE, reporting that top-ranked SNPs yield an average AUC of 0.671, outperforming bottom-ranked (0.497) and randomly selected (0.558) SNPs. They also report that the identified top SNPs include known AD-associated variants, and they make the code and a web server publicly available.

Significance. If the reported results held, the paper would offer a useful, interpretable SNP-prioritization tool for AD and a general architecture for integrating local and global genomic information. The open-source implementation, the identification of known APOE variants (rs429358, rs7412), and the explicit attempt to combine two complementary modeling strategies are strengths. However, the central performance claim is currently not supported: the evaluation is circular (SNP ranking and validation share the same samples), α is tuned on the same data, no error bars or external validation are provided, and no APOE-only baseline is reported. The significance of the framework therefore hinges on a re-analysis that corrects these biases.

major comments (4)
  1. [Model Evaluation and SNP Prioritization] The central claim of the paper rests on a circular evaluation. After selecting the optimal α, the authors assign each SNP a combined score computed from out-of-fold predictions generated on the full 1,050-sample dataset, then rank all SNPs by that score and use the same 1,050 samples in a fresh five-fold cross-validation to evaluate the predictive AUC of the resulting top-, bottom-, and random-ranked SNP panels. Because the ranking already incorporates information derived from every sample's label (via the out-of-fold predictions), the top-ranked SNPs are selected using test-label information. The reported AUCs in Table 2 (0.697, 0.660, 0.657) therefore do not measure the ability of the model to rank SNPs in an unbiased fashion, and the 35.0%/20.3% improvements over bottom/random panels are inflated. The evaluation must be carried out in a fully nested scheme: within each training fold, compute SNP scores and select α using only that fold's training data, then evaluate the selected panel on the held-out fold.
  2. [Combining Local and Global Accuracy] The weighting parameter α is itself fitted on the same data used for final evaluation. For each candidate α, the top 100 SNPs under that weighting are selected and a stacked classifier is evaluated via cross-validation on the same 1,050 samples; the α with the highest average AUC is then used to produce the final results reported in Table 2 and Figure 2. This makes α a fitted quantity rather than a fixed hyperparameter, and the final performance estimate is therefore optimistic. The α selection must be nested inside the training folds (or performed on a separate validation set) to produce an unbiased estimate.
  3. [Results / Table 2] No measure of uncertainty is provided for any of the reported AUCs. With 1,050 samples and a feature selection step that involves ranking 14,011 SNPs, the differences in Table 2 (e.g., 0.697 vs 0.456 for the 100-SNP panel) could be unstable across random fold partitions. The authors should report confidence intervals, standard errors, or results from repeated cross-validation, and test whether the top-ranked AUC is significantly greater than the random/bottom AUC under a paired test.
  4. [Study participants / Whole-genome sequencing data] The analysis is restricted to a 1-Mb region surrounding APOE, and the top-100 ranked SNPs are reported to come almost entirely from APOE and APOC1. Since APOE ε4 carrier status is the strongest known genetic risk factor for AD, the authors must include an APOE-only baseline (e.g., a logistic regression model using rs429358/rs7412 genotype dosage) to demonstrate that DuAL-Net's SNP prioritization adds predictive value beyond what is already captured by APOE. Without this control, the clinical and scientific significance of the 0.671 AUC is unclear.
minor comments (4)
  1. [Local SNP Window Analysis] The results section states that 'the genome was divided into 140 contiguous windows of 100 SNPs each,' conflicting with the Methods description of 'non-overlapping 100-base pair sized windows'; please clarify whether windows are defined by SNP count or base-pair coordinates.
  2. [References] There is a typo in the reference section heading: 'Referecnes' should be 'References'.
  3. [References] Reference 14 is incomplete; it should include the full citation for TabNet (Arik and Pfister, 2019).
  4. [SNP Ranking and ROC Performance] Supplementary Table 1 is referenced in the text but is not available in the manuscript; if submitted as a separate file, it should be mentioned in the main text.

Circularity Check

3 steps flagged · score 8.0 of 10

The top-ranked SNP AUC is produced by the same cross-validated selection loop used to choose α and the ranking, all on the same 1,050 samples; the central improvement claim reduces to feature-selection bias.

  1. fitted input called prediction [Methods, 'Combining Local and Global Accuracy' (α optimization)]
    "For each candidate α, all SNPs were re-ranked according to their combined score, and the top 100 SNPs under that weighting were selected. A stacked classifier, restricted to these 100 SNPs, was then trained and evaluated via cross-validation to determine its ability to distinguish AD cases from controls. The α that produced the highest average AUC across folds was selected as the best trade-off between local and global information."

    The reported top-100 AUC (0.697) is produced by the same CV-selection loop used to pick α and the top-100 panel. The loop 'choose α by CV top-100 AUC; then report CV top-100 AUC as a result' is closed on the same 1,050 samples and labels, so the headline result is the fitted quantity itself, not an independent prediction of the selected panel's performance.

  2. fitted input called prediction [Methods, 'Model Evaluation and SNP Prioritization']
    "After selecting the optimal α, all SNPs were assigned final combined scores and ranked from most to least likely to be associated with AD. To evaluate the methodology, we created multiple SNP panels of varying sizes (100, 500, and 1000) from the top-ranked SNPs. For comparison, we also created corresponding panels of the same sizes using the bottom-ranked SNPs and randomly selected SNPs. Classification models were trained on these respective sets using a five-fold cross-validation framework."

    The combined scores used for ranking are functions of OOF predictions and annotation-group accuracies computed on the full 1,050-sample dataset, so the top-ranked panel is selected using all sample labels. The 'evaluation' then trains and cross-validates on the same 1,050 samples. The 0.671 average top-ranked AUC and the 35.0%/20.3% improvements over bottom/random panels therefore measure label-informed selection on the labels that generated the ranking, i.e., feature-selection bias rather than a valid prioritization result.

1 more flagged steps
  1. self definitional [Methods, 'Global Annotation-Based Modeling']
    "Each SNP then received a global accuracy score calculated by averaging the classification accuracies of all annotation groups in which it appeared."

    The global SNP score used for ranking is, by definition, the average classification accuracy of models trained on the same samples' AD/CN labels. A SNP's importance is thus defined as its group's in-sample accuracy, and the ranking built from these scores is then validated on the same labels. This component of the combined score is a fitted training statistic repackaged as a predictive importance measure.

full rationale

The circularity is not in the self-citation to the authors' earlier imputation work (ref. 22), which is a preprocessing detail and does not carry the central claim; it is in the evaluation loop. In 'Combining Local and Global Accuracy', α is selected by training a stacked classifier on the top-100 SNPs ranked by Combined score = α × local accuracy score + (1−α) × global accuracy score and evaluating that classifier by CV on the same 1,050 samples. The local and global scores are themselves accuracies of models fitted on those same samples. Then in 'Model Evaluation and SNP Prioritization', the same combined scores are used to rank all SNPs and the top/bottom/random panels are evaluated in a five-fold CV on the same samples. Consequently, the reported top-100 AUC of 0.697 and average top-ranked AUC of 0.671 are the selection criterion and label-informed selection results, not unbiased estimates of how the model would rank SNPs on unseen data. Because no held-out set is used for SNP ranking, α selection, or panel evaluation, the headline improvement over bottom-ranked and random SNPs is forced by using the test labels during feature selection. The only external support is biological plausibility (known AD SNPs like rs429358/rs7412 are ranked highly), which is a sanity check and does not validate the quantitative AUC claim. I therefore score the paper 8: the central prediction claim reduces by construction to the fitted ranking criterion.

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

The framework itself is the contribution; it introduces no new biological entities. The main free choices are the weighting alpha, window size, panel sizes, annotation grouping, and unstated hyperparameters.

free parameters (5)
  • alpha (α) = 0.8
    Combination weight for local vs global scores; selected by five-fold CV to maximize top-100 SNP panel AUC on the same 1,050 samples, so the reported top-100 AUC depends directly on this fitted choice.
  • Window size = 100 SNPs per window (text also says 100 base pairs)
    The genome is segmented into 140 windows of 100 SNPs each; the choice of window size is arbitrary and affects local scores.
  • SNP subset sizes = 100, 500, 1000
    Evaluation panel sizes chosen by the authors; results vary by subset size.
  • Annotation categories = hand-selected Ensembl categories (genomic context, biotype, consequence, clinical significance)
    The set of annotation groups is chosen ad hoc and determines each SNP's global score; no sensitivity analysis is provided.
  • TabNet/RF hyperparameters = not specified
    Model hyperparameters are not reported, so the fitted values are unknown but are free choices affecting results.
assumptions (4)
  • domain assumption The 1 Mb region around APOE on chromosome 19 contains enough predictive signal for AD dementia to evaluate the framework.
    The analysis is restricted to this region (Methods, Whole-genome sequencing data); results for whole-genome prediction are not established.
  • domain assumption No population stratification or cryptic relatedness confounds SNP-disease associations in the ADNI/ADSP sample.
    No genetic ancestry principal components or relatedness adjustment is described beyond Pihat > 0.4 exclusion (Methods, Whole-genome sequencing data).
  • standard math Out-of-fold stacking provides unbiased base-model predictions for downstream SNP ranking.
    Standard stacked generalization assumption, but downstream feature selection on the same data violates the unbiasedness premise (Methods, Model Evaluation).
  • domain assumption Ensembl annotation categories are complete and correctly mapped for all SNPs.
    Relying on Ensembl annotations may bias toward well-known loci, as the authors themselves note in the Discussion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DuAL-Net: A Hybrid Framework for Alzheimer's Disease Prediction from Whole-Genome Sequencing via Local SNP Windows and Global Annotations." pith.science (2026). https://pith.science/paper/5BJSX4Y2

@misc{pith2026250600673,
  author       = {Pith},
  title        = {Pith review of: DuAL-Net: A Hybrid Framework for Alzheimer's Disease Prediction from Whole-Genome Sequencing via Local SNP Windows and Global Annotations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5BJSX4Y2}},
  note         = {Machine review of arXiv:2506.00673}
}
read the original abstract

Alzheimer's disease (AD) dementia is the most common form of dementia. With the emergence of disease-modifying therapies, predicting disease risk before symptom onset has become critical. We introduce DuAL-Net, a hybrid deep learning framework for AD dementia prediction using whole genome sequencing (WGS) data. DuAL-Net integrates two components: local probability modeling, which segments the genome into non-overlapping windows, and global annotation-based modeling, which annotates SNPs and reorganizes WGS input to capture long-range functional relationships. Both employ out-of-fold stacking with TabNet and Random Forest classifiers. Final predictions combine local and global probabilities using an optimized weighting parameter alpha. We analyzed WGS data from 1,050 individuals (443 cognitively normal, 607 AD dementia) using five-fold cross-validation. DuAL-Net achieved an AUC of 0.671 using top-ranked SNPs, representing 35.0% and 20.3% higher performance than bottom-ranked and randomly selected SNPs, respectively. ROC analysis demonstrated strong positive correlation between SNP prioritization rank and predictive power. The model identified known AD-associated SNPs as top contributors alongside potentially novel variants. DuAL-Net presents a promising framework improving both predictive accuracy and biological interpretability. The framework and web implementation offer an accessible platform for broader research applications.

Figures

Figures reproduced from arXiv: 2506.00673 by the authors.

Figure 2
Figure 2. ROC curves comparing top-ranked, bottom-ranked and randomly selected SNP subsets identified by DuAL-Net To evaluate the discriminative power of the ranking calculated by DuAL-Net, we compared ROC curves among three groups (top-ranked SNPs, bottom-ranked SNPs, randomly selected SNPs) within different subset sizes (100, 500 and 1000 SNPs). The top-ranked SNPs presented higher AUC values than bottom￾ranked or randomly … view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages

  1. [1]

    Global action plan on the public health response to dementia 2017–2025

    Organization WH. Global action plan on the public health response to dementia 2017–2025. Geneva: World Health Organization, 2017

    work page 2017

  2. [2]

    Antiamyloid Monoclonal Antibody Therapy for Alzheimer Disease: Emerging Issues in Neurology

    Ramanan VK, Armstrong MJ, Choudhury P, et al. Antiamyloid Monoclonal Antibody Therapy for Alzheimer Disease: Emerging Issues in Neurology. Neurology 2023;101:842-852

    work page 2023

  3. [3]

    Donanemab in Early Symptomatic Alzheimer Disease: The TRAILBLAZER-ALZ 2 Randomized Clinical Trial

    Sims JR, Zimmer JA, Evans CD, et al. Donanemab in Early Symptomatic Alzheimer Disease: The TRAILBLAZER-ALZ 2 Randomized Clinical Trial. JAMA 2023

    work page 2023

  4. [4]

    Role of genes and environments for explaining Alzheimer disease

    Gatz M, Reynolds CA, Fratiglioni L, et al. Role of genes and environments for explaining Alzheimer disease. Arch Gen Psychiatry 2006;63:168-174

    work page 2006

  5. [5]

    Heritability for Alzheimer's disease: the study of dementia in Swedish twins

    Gatz M, Pedersen NL, Berg S, et al. Heritability for Alzheimer's disease: the study of dementia in Swedish twins. J Gerontol A Biol Sci Med Sci 1997;52:M117-125

    work page 1997

  6. [6]

    A global reference for human genetic variation

    Genomes Project C, Auton A, Brooks LD, et al. A global reference for human genetic variation. Nature 2015;526:68-74

    work page 2015

  7. [7]

    Genomic Analysis in the Age of Human Genome Sequencing

    Lappalainen T, Scott AJ, Brandt M, Hall IM. Genomic Analysis in the Age of Human Genome Sequencing. Cell 2019;177:70-84

    work page 2019

  8. [8]

    Machine learning for genetic prediction of psychiatric disorders: a systematic review

    Bracher-Smith M, Crawford K, Escott -Price V. Machine learning for genetic prediction of psychiatric disorders: a systematic review. Mol Psychiatry 2021;26:70-79

    work page 2021

Show all 43 references
  1. [9]

    Zaranek SW, Zaranek AW, Amstutz P, et al. Discovering Genetic Signatures Associated with Alzheimer’s Disease in Tiled Whole Genome Sequence Data: Results from the Artificial Intelligence for Alzheimer’s Disease (AI4AD) Consortium. 2024

    2024

  2. [10]

    Prediction of antimicrobial resistance based on whole - genome sequencing and machine learning

    Ren Y, Chakraborty T, Doijad S, et al. Prediction of antimicrobial resistance based on whole - genome sequencing and machine learning. Bioinformatics 2022;38:325-334

    2022

  3. [11]

    Machine learning random forest for predicting oncosomatic variant NGS analysis

    Pellegrino E, Jacques C, Beaufils N, et al. Machine learning random forest for predicting oncosomatic variant NGS analysis. Sci Rep 2021;11:21820

    2021

  4. [12]

    Classifying Cancer Patients Based on DNA Sequences Using Machine Learning

    Hussain F, Saeed U, Muhammad G, Islam N, Sheikh GS. Classifying Cancer Patients Based on DNA Sequences Using Machine Learning. Journal of Medical Imaging and Health Informatics 2019;9:436-443

    2019

  5. [13]

    Classification of Alzheimer's disease using robust TabNet neural networks on genetic data

    Jin Y, Ren Z, Wang W, et al. Classification of Alzheimer's disease using robust TabNet neural networks on genetic data. Math Biosci Eng 2023;20:8358-8374

    2023

  6. [14]

    Sercan O. Arik TP. TabNet: Attentive Interpretable Tabular Learning

  7. [15]

    Genetic studies of quantitative MCI and AD phenotypes in ADNI: Progress, opportunities, and plans

    Saykin AJ, Shen L, Yao X, et al. Genetic studies of quantitative MCI and AD phenotypes in ADNI: Progress, opportunities, and plans. Alzheimers Dement 2015;11:792-814

    2015

  8. [16]

    An interpretable Alzheimer's disease oligogenic risk score informed by neuroimaging biomarkers improves risk prediction and stratification

    Suh EH, Lee G, Jung SH, et al. An interpretable Alzheimer's disease oligogenic risk score informed by neuroimaging biomarkers improves risk prediction and stratification. Front Aging Neurosci 2023;15:1281748

    2023

  9. [17]

    The Alzheimer's Disease Sequencing Project: Study design and sample selection

    Beecham GW, Bis JC, Martin ER, et al. The Alzheimer's Disease Sequencing Project: Study design and sample selection. Neurol Genet 2017;3:e194

    2017

  10. [18]

    Fast and accurate short read alignment with Burrows -Wheeler transform

    Li H, Durbin R. Fast and accurate short read alignment with Burrows -Wheeler transform. Bioinformatics 2009;25:1754-1760

    2009

  11. [19]

    The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data

    McKenna A, Hanna M, Banks E, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res 2010;20:1297-1303

    2010

  12. [20]

    VCPA: genomic variant calling pipeline and data management tool for Alzheimer's Disease Sequencing Project

    Leung YY, Valladares O, Chou YF, et al. VCPA: genomic variant calling pipeline and data management tool for Alzheimer's Disease Sequencing Project. Bioinformatics 2019;35:1768 -1770

    2019

  13. [21]

    NIAGADS Alzheimer's GenomicsDB: A resource for exploring Alzheimer's disease genetic and genomic knowledge

    Greenfest-Allen E, Valladares O, Kuksa PP, et al. NIAGADS Alzheimer's GenomicsDB: A resource for exploring Alzheimer's disease genetic and genomic knowledge. Alzheimers Dement 2024;20:1123 - 1136

    2024

  14. [22]

    LD -informed deep learning for Alzheimer's gene loci detection using WGS data

    Jo T, Bice P, Nho K, Saykin AJ, Alzheimer's Disease Sequencing P. LD -informed deep learning for Alzheimer's gene loci detection using WGS data. Alzheimers Dement (N Y) 2025;11:e70041

    2025

  15. [23]

    Cross-attention enables deep learning on limited omics- imaging-clinical data of 130 lung cancer patients

    Verma S, Magazzu G, Eftekhari N, et al. Cross-attention enables deep learning on limited omics- imaging-clinical data of 130 lung cancer patients. Cell Rep Methods 2024;4:100817

    2024

  16. [24]

    ST-Align: A Multimodal Foundation Model for Image-Gene Alignment in Spatial Transcriptomics

    Yuxiang Lin LL, Ying Chen, Xushi Zhang, Zihui Wang, Wenxian Yang, Mengsha Tong, Rongshan Yu. ST-Align: A Multimodal Foundation Model for Image-Gene Alignment in Spatial Transcriptomics

  17. [25]

    Discovering Alzheimer Genetic Biomarkers Using Bayesian Networks

    Sherif FF, Zayed N, Fakhr M. Discovering Alzheimer Genetic Biomarkers Using Bayesian Networks. Adv Bioinformatics 2015;2015:639367

    2015

  18. [26]

    Early detection of Alzheimer's disease using single nucleotide polymorphisms analysis based on gradient boosting tree

    Ahmed H, Soliman H, Elmogy M. Early detection of Alzheimer's disease using single nucleotide polymorphisms analysis based on gradient boosting tree. Comput Biol Med 2022;146:105622

    2022

  19. [27]

    Genome -wide association analysis of dementia and its clinical endophenotypes reveal novel loci associated with Alzheimer's disease and three causality networks: The GR@ACE project

    Moreno-Grau S, de Rojas I, Hernandez I, et al. Genome -wide association analysis of dementia and its clinical endophenotypes reveal novel loci associated with Alzheimer's disease and three causality networks: The GR@ACE project. Alzheimers Dement 2019;15:1333-1347

    2019

  20. [28]

    Genome -wide meta -analysis, fine -mapping and integrative prioritization implicate new Alzheimer's disease risk genes

    Schwartzentruber J, Cooper S, Liu JZ, et al. Genome -wide meta -analysis, fine -mapping and integrative prioritization implicate new Alzheimer's disease risk genes. Nat Genet 2021;53:392 -402

    2021

  21. [29]

    Sex-specific genetic architecture of late-life memory performance

    Eissman JM, Archer DB, Mukherjee S, et al. Sex-specific genetic architecture of late-life memory performance. Alzheimers Dement 2024;20:1250-1267

    2024

  22. [30]

    Similar Genetic Architecture of Alzheimer's Disease and Differential APOE Effect Between Sexes

    Wang H, Lo MT, Rosenthal SB, et al. Similar Genetic Architecture of Alzheimer's Disease and Differential APOE Effect Between Sexes. Front Aging Neurosci 2021;13:674318

    2021

  23. [31]

    A novel mutation in the apolipoprotein E gene (APOE*4 Pittsburgh) is associated with the risk of late -onset Alzheimer's disease

    Kamboh MI, Aston CE, Perez-Tur J, et al. A novel mutation in the apolipoprotein E gene (APOE*4 Pittsburgh) is associated with the risk of late -onset Alzheimer's disease. Neurosci Lett 1999;263:129 - 132

    1999

  24. [32]

    Apolipoprotein E Pittsburgh variant is not associated with the risk of late -onset Alzheimer's disease in a Spanish population

    Baron M, Jimenez-Escrig A, Orensanz L, Simon J, Perez-Tur J. Apolipoprotein E Pittsburgh variant is not associated with the risk of late -onset Alzheimer's disease in a Spanish population. Am J Med Genet B Neuropsychiatr Genet 2003;120B:121-124

    2003

  25. [33]

    The contribution of common and rare genetic variants to variation in metabolic traits in 288,137 East Asians

    Kim YJ, Moon S, Hwang MY, et al. The contribution of common and rare genetic variants to variation in metabolic traits in 288,137 East Asians. Nat Commun 2022;13:6642

    2022

  26. [34]

    Characterising metabolomic signatures of lipid - modifying therapies through drug target mendelian randomisation

    Richardson TG, Leyden GM, Wang Q, et al. Characterising metabolomic signatures of lipid - modifying therapies through drug target mendelian randomisation. PLoS Biol 2022;20:e3001547

    2022

  27. [35]

    Genetics of 35 blood and urine biomarkers in the UK Biobank

    Sinnott-Armstrong N, Tanigawa Y, Amar D, et al. Genetics of 35 blood and urine biomarkers in the UK Biobank. Nat Genet 2021;53:185-194

    2021

  28. [36]

    Multi -trait discovery and fine -mapping of lipid loci in 125,000 individuals of African ancestry

    Kamiza AB, Toure SM, Zhou F, et al. Multi -trait discovery and fine -mapping of lipid loci in 125,000 individuals of African ancestry. Nat Commun 2023;14:5403

    2023

  29. [37]

    Minority -centric meta-analyses of blood lipid levels identify novel loci in the Population Architecture using Genomics and Epidemiology (PAGE) study

    Hu Y, Graff M, Haessler J, et al. Minority -centric meta-analyses of blood lipid levels identify novel loci in the Population Architecture using Genomics and Epidemiology (PAGE) study. PLoS Genet 2020;16:e1008684

    2020

  30. [38]

    The power of genetic diversity in genome-wide association studies of lipids

    Graham SE, Clarke SL, Wu KH, et al. The power of genetic diversity in genome-wide association studies of lipids. Nature 2021;600:675-679

    2021

  31. [39]

    Pleiotropic genetic architecture and novel loci for C - reactive protein levels

    Koskeridis F, Evangelou E, Said S, et al. Pleiotropic genetic architecture and novel loci for C - reactive protein levels. Nat Commun 2022;13:6939

    2022

  32. [40]

    Impact of hs -CRP concentration on brain structure alterations and cognitive trajectory in Alzheimer's disease

    Zhang Y, Tatewaki Y, Nakase T, et al. Impact of hs -CRP concentration on brain structure alterations and cognitive trajectory in Alzheimer's disease. Front Aging Neurosci 2023;15:1227325

    2023

  33. [41]

    Serum high -sensitivity C-reactive protein and dementia in a community-dwelling Japanese older population (JPSC-AD)

    Tachibana A, Iga JI, Ozaki T, et al. Serum high -sensitivity C-reactive protein and dementia in a community-dwelling Japanese older population (JPSC-AD). Sci Rep 2024;14:7374

    2024

  34. [42]

    Chronic kidney disease: a risk factor for dementia onset: a population -based study

    Sasaki Y, Marioni R, Kasai M, Ishii H, Yamaguchi S, Meguro K. Chronic kidney disease: a risk factor for dementia onset: a population -based study. The Osaki -Tajiri Project. J Am Geriatr Soc 2011;59:1175-1181

    2011

  35. [43]

    Kidney Function, Kidney Function Decline, and the Risk of Dementia in Older Adults: A Registry-Based Study

    Xu H, Garcia-Ptacek S, Trevisan M, et al. Kidney Function, Kidney Function Decline, and the Risk of Dementia in Older Adults: A Registry-Based Study. Neurology 2021;96:e2956-e2965

    2021

Pith tools

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