hdWGCNA identifies co-expression networks in high-dimensional transcriptomics data

In the paper

“Co-expression networks are based on transformed pairwise correlations of input features, resulting in a quantitative measure of relatedness between genes.1,2”

What the sources say

• Supported

  The Background glossary defines co-expression networks as weighted gene networks whose edges are determined by pairwise correlations of gene-expression profiles. In Results, “Category 1: Functions for network construction” explains that correlation-based similarity is transformed into adjacency, a continuous measure of connection strength between genes. (LLM, conf=0.94)

  cites [1] Langfelder P. Horvath S. WGCNA: an R package for weighted correlation network analysis BMC Bioinf. 9 2008 559 10.1186/1471-2105-9-559 PMC2631488 19114008 source
• Supported

  The Methods section (‘Yeast gene co-expression network construction’) says the authors calculated pairwise Pearson correlations between genes and thresholded their absolute values to construct the network. The Results section (‘Comparing GTOMm to the correlation coefficient’) also defines a quantitative, correlation-based gene-pair dissimilarity measure. Together, these support the attributed description of transformed correlations as a measure of relatedness. (LLM, conf=0.88)

  cites [2] Yip A.M. Horvath S. Gene network interconnectedness and the generalized topological overlap measure BMC Bioinf. 8 2007 22 10.1186/1471-2105-8-22 PMC1797055 17250769 source

In the paper

“Hierarchical clustering on the network structure allows us to uncover functional modules of genes whose expression profiles are tightly intertwined,3,4 which typically correspond to specific biological processes and disease states.”

What the sources say

• ?Not checked — no full text of the sourcecites [3] Langfelder P. Zhang B. Horvath S. Defining clusters from a hierarchical cluster tree: the Dynamic Tree Cut package for R Bioinformatics 24 2008 719 720 10.1093/bioinformatics/btm563 18024473 source
• Partly supported

  The Background section describes defining modules as branches of a hierarchical-clustering dendrogram, and the Results section applies this method to gene co-expression networks. The Results also report that most modules are enriched for known gene ontologies, supporting a link to biological functions. The paper does not establish that these modules typically correspond to disease states. (LLM, conf=0.87)

  cites [4] Dong J. Horvath S. Understanding network concepts in modules BMC Syst. Biol. 1 2007 24 10.1186/1752-0509-1-24 17547772 PMC3238286 sourceExperimental — accuracy not yet measured

In the paper

“Optionally, ModuleEigengenes can employ the harmony22 algorithm following SVD, which increases runtime but not memory usage.”

What the sources say

• Partly supported

  The Online Methods > Harmony section (page not provided) describes Harmony operating on a PCA embedding, supporting its use after dimensionality reduction. The Results > Harmony Scales to Large Data section reports Harmony’s runtime and memory usage, but does not assess its use in ModuleEigengenes or show that adding it increases runtime without increasing memory usage. (LLM, conf=0.85)

  cites [22] Korsunsky I. Millard N. Fan J. Slowikowski K. Zhang F. Wei K. Baglaenko Y. Brenner M. Loh P.-r. Raychaudhuri S. Fast, sensitive and accurate integration of single-cell data with Harmony Nat. Methods 16 2019 1289 1296 10.1038/s41592-019-0619-0 31740819 PMC6884693 sourceExperimental — accuracy not yet measured

In the paper

“the hdWGCNA R package extends the data structures and functionality of the widely used Seurat package,5,6,7”

What the sources say

• Supported

  The citations follow “Seurat package,” so this source is used to identify Seurat, not to establish what hdWGCNA extends. The Results section (Overview of Seurat alignment workflow; page unavailable) describes Seurat’s workflow and functionality, supporting that identification. (LLM, conf=0.89)

  cites [5] Butler A. Hoffman P. Smibert P. Papalexi E. Satija R. Integrating single-cell transcriptomic data across different conditions, technologies, and species Nat. Biotechnol. 36 2018 411 420 10.1038/nbt.4096 29608179 PMC6700744 source
• Supported

  The citation identifies Seurat, the package named in the citing sentence. The paper’s Data and software availability section identifies Seurat v3 as a CRAN package, and its Method details describes Seurat’s functions. The statement about what hdWGCNA extends is the citing authors’ description of their own package. (LLM, conf=0.88)

  cites [6] Stuart T. Butler A. Hoffman P. Hafemeister C. Papalexi E. Mauck W.M. Hao Y. Stoeckius M. Smibert P. Satija R. Comprehensive integration of single-cell data Cell 177 2019 1888 1902.e21 10.1016/j.cell.2019.05.031 31178118 PMC6687398 source
• Supported

  The citation is attached to the identification of Seurat, rather than to the citing authors’ statement about what hdWGCNA extends. The STAR★Methods key resources table lists Seurat v4 as software from this paper, and Method details identify it as an R package used for single-cell analysis. The source need not describe hdWGCNA to support this identification. (LLM, conf=0.90)

  cites [7] Hao Y. Hao S. Andersen-Nissen E. Mauck W.M. Zheng S. Butler A. Lee M.J. Wilk A.J. Darby C. Integrated analysis of multimodal single-cell data Cell 184 2021 3573 3587.e29 10.1016/j.cell.2021.04.048 34062119 PMC8238499 source

In the paper

“To maximize usability among the genomics community, the hdWGCNA R package extends the data structures and functionality of the widely used Seurat package,5,6,7 and we developed an extensive documentation website for hdWGCNA demonstrating its use on new datasets.”

What the sources say

• Supported

  The citation is attached to “Seurat package,” not to the citing authors’ statements about developing hdWGCNA or its documentation. The Discussion (page unavailable in our copy) identifies Seurat as an R toolkit in which the paper’s procedure was implemented, supporting the identification of the package. (LLM, conf=0.91)

  cites [5] Butler A. Hoffman P. Smibert P. Papalexi E. Satija R. Integrating single-cell transcriptomic data across different conditions, technologies, and species Nat. Biotechnol. 36 2018 411 420 10.1038/nbt.4096 29608179 PMC6700744 source
• Supported

  The citation is attached to the Seurat package, not to the citing authors’ statements about developing hdWGCNA or its documentation website. The cited paper’s STAR Methods (Data and software availability) identify Seurat v3 as the package implementing its methods and give its CRAN and GitHub locations. (LLM, conf=0.91)

  cites [6] Stuart T. Butler A. Hoffman P. Hafemeister C. Papalexi E. Mauck W.M. Hao Y. Stoeckius M. Smibert P. Satija R. Comprehensive integration of single-cell data Cell 177 2019 1888 1902.e21 10.1016/j.cell.2019.05.031 31178118 PMC6687398 source
• Supported

  The citations are attached to the identification of Seurat, not to the citing authors’ statements about developing hdWGCNA or its website. The Introduction describes Seurat as an open-source R toolkit, and the STAR Methods key resources table lists Seurat v4 as software from this paper (page not provided). (LLM, conf=0.90)

  cites [7] Hao Y. Hao S. Andersen-Nissen E. Mauck W.M. Zheng S. Butler A. Lee M.J. Wilk A.J. Darby C. Integrated analysis of multimodal single-cell data Cell 184 2021 3573 3587.e29 10.1016/j.cell.2021.04.048 34062119 PMC8238499 source

In the paper

“Cd9 encodes a transmembrane protein and is a known glioblastoma biomarker,40”

What the sources say

• Supported

  The Abstract states that CD9 encodes a transmembrane protein and reports its up-regulation in glioblastoma tissue. The Introduction describes CD9 as a biomarker evaluated for glioblastoma stem cells, supporting the citing sentence's characterization. (LLM, conf=0.92)

  cites [40] Podergajs N. Motaln H. Rajčević U. Verbovšek U. Koršič M. Obad N. Espedal H. Vittori M. Herold-Mende C. Miletic H. Transmembrane protein CD9 is glioblastoma biomarker, relevant for maintenance of glioblastoma stem cells Oncotarget 7 2016 593 609 10.18632/oncotarget.5477 26573230 PMC4808020 source

In the paper

“Radial glia, which share transcriptomic similarities with mature astrocytes, are progenitor cells that give rise to numerous distinct cell fates, including neuronal cells, astrocytes, oligodendrocytes, and ependymal cells.34,35”

What the sources say

• Partly supported

  The Results section, “Extensive Neurogenic Capacity of oRG Cells” (page unavailable), reports that outer radial glia share some genes with later astrocytes and that single cells produced neuronal and glial daughters. It does not establish broad transcriptomic similarity to mature astrocytes or identify astrocytes, oligodendrocytes, and ependymal cells as distinct daughter-cell fates. (LLM, conf=0.87)

  cites [34] Pollen A.A. Nowakowski T.J. Chen J. Retallack H. Sandoval-Espinosa C. Nicholas C.R. Shuga J. Liu S.J. Oldham M.C. Diaz A. Molecular identity of human outer radial glia during cortical development Cell 163 2015 55 67 10.1016/j.cell.2015.09.004 26406371 PMC4583716 sourceExperimental — accuracy not yet measured
• Partly supported

  The Results (RNA-Velocity Analysis of Lineage; page unavailable) describe radial glia giving rise to neurons, and the Discussion (page unavailable) describes progenitors giving rise to neuronal and glial diversity. Our copy does not establish transcriptomic similarity between radial glia and mature astrocytes or identify astrocytes, oligodendrocytes, and ependymal cells as the specific fates claimed. (LLM, conf=0.85)

  cites [35] Eze U.C. Bhaduri A. Haeussler M. Nowakowski T.J. Kriegstein A.R. Single-cell atlas of early human brain development highlights heterogeneity of human neuroepithelial cells and early radial glia Nat. Neurosci. 24 2021 584 594 10.1038/s41593-020-00794-1 33723434 PMC8012207 sourceExperimental — accuracy not yet measured

In the paper

“ST enables the investigation of biological patterns that might otherwise be hidden in other -omics technologies, such as scRNA-seq or bulk RNA-seq.26,27”

What the sources say

• Supported

  The Data analysis section says spatial transcriptomics enables identification of spatially variable genes and inference of cell–cell interactions. The Upstream section explains that spatial information enables analysis of neighboring cells and spatial gene patterns, supporting the claim that these patterns may be missed without spatial measurements. No page numbers are available in our copy. (LLM, conf=0.87)

  cites [26] Moses L. Pachter L. Museum of Spatial Transcriptomics Preprint at bioRxiv 2021 10.1101/2021.05.11.443152 35273392 source
• ?Not checked — no full text of the sourcecites [27] Moffitt J.R. Lundberg E. Heyn H. The emerging landscape of spatial profiling technologies Nat. Rev. Genet. 23 2022 741 759 10.1038/s41576-022-00515-3 35859028 source

In the paper

“One of the INH-M13 hub genes is CHD2, whose de novo variants have been identified in individuals with ASD.42,43”

What the sources say

• Supported

  The Introductory section states that a de novo CHD2 missense mutation was reported in an individual with autism spectrum disorder. The paper presents this earlier finding as evidence that CHD2 mutations occur across neurodevelopmental disorders; the claim does not attribute that identification to this paper's own cohort. (LLM, conf=0.92)

  cites [42] Carvill G.L. Heavin S.B. Yendle S.C. McMahon J.M. O’Roak B.J. Cook J. Khan A. Dorschner M.O. Weaver M. Calvert S. Targeted resequencing in epileptic encephalopathies identifies de novo mutations in CHD2 and SYNGAP1 Nat. Genet. 45 2013 825 830 10.1038/ng.2646 23708187 PMC3704157 source
• ?Could not verify

  The Abstract (page unavailable) reports de novo variants in ASD cases and highlights CHD8, but it does not identify a CHD2 variant. Searches of our copy for “CHD2” and “chromodomain helicase DNA binding protein 2” found no passage establishing the gene-specific claim. (LLM, conf=0.73)

  cites [43] Neale B.M. Kou Y. Liu L. Ma’ayan A. Samocha K.E. Sabo A. Lin C.-F. Stevens C. Wang L.-S. Makarov V. Patterns and rates of exonic de novo mutations in autism spectrum disorders Nature 485 2012 242 245 10.1038/nature11011 22495311 PMC3613847 sourceExperimental — accuracy not yet measured

In the paper

“Microglia, the resident immune cells of the brain, are implicated in the pathology and genetic risk of several CNS diseases, including AD.46,47,48,49”

What the sources say

• Partly supported

  The Results section (no page provided) reports suggestive associations between microglial gene expression and Alzheimer’s disease risk, and the Discussion links microglia to a hypothesis about AD pathogenesis. The study does not establish the same implication for several other CNS diseases; its microglia associations also did not survive the stringent Bonferroni correction. (LLM, conf=0.85)

  cites [46] Jansen I.E. Savage J.E. Watanabe K. Bryois J. Williams D.M. Steinberg S. Sealock J. Karlsson I.K. Hägg S. Athanasiu L. Genome-wide meta-analysis identifies new loci and functional pathways influencing Alzheimer’s disease risk Nat. Genet. 51 2019 404 413 10.1038/s41588-018-0311-9 30617256 PMC6836675 sourceExperimental — accuracy not yet measured
• Partly supported

  The Results section (Candidate gene prioritization at genome-wide loci; page unavailable) links microglial gene expression to Alzheimer’s disease risk loci and describes microglia as involved in the pathological immune response in late-onset Alzheimer’s disease. The paper does not establish the same link for several CNS diseases, so the claim is supported for Alzheimer’s disease but not for the broader group. (LLM, conf=0.87)

  cites [47] Kunkle B.W. Grenier-Boley B. Sims R. Bis J.C. Damotte V. Naj A.C. Boland A. Vronskaya M. van der Lee S.J. Amlie-Wolf A. Genetic meta-analysis of diagnosed Alzheimer’s disease identifies new risk loci and implicates A, tau, immunity and lipid processing Nat. Genet. 51 2019 414 430 10.1038/s41588-019-0358-2 30820047 PMC6463297 sourceExperimental — accuracy not yet measured
• Partly supported

  The Results section (‘AD risk is enriched near genes with high microglial gene expression’; page not provided) reports enrichment of Alzheimer’s disease risk near genes highly expressed in microglia. The paper also discusses a possible role for microglial activation in AD, but the passages do not extend that finding to several CNS diseases. (LLM, conf=0.88)

  cites [48] Schwartzentruber J. Cooper S. Liu J.Z. Barrio-Hernandez I. Bello E. Kumasaka N. Young A.M.H. Franklin R.J.M. Johnson T. Estrada K. Genome-wide meta-analysis, fine-mapping and integrative prioritization implicate new Alzheimer’s disease risk genes Nat. Genet. 53 2021 392 402 10.1038/s41588-020-00776-w 33589840 PMC7610386 sourceExperimental — accuracy not yet measured

In the paper

“Similarly, rare loss-of-function mutations have been reported in the SCN2A gene, a hub gene of the INH-M11 module.44”

What the sources say

• Supported

  The Results (Primary Analysis) identify a novel nonsense variant in SCN2A, and the Abstract (Conclusion) describes very rare loss-of-function mutations at this gene. These passages support the attributed claim that rare SCN2A loss-of-function mutations have been reported. (LLM, conf=0.94)

  cites [44] Carroll L.S. Woolf R. Ibrahim Y. Williams H.J. Dwyer S. Walters J. Kirov G. O’Donovan M.C. Owen M.J. Mutation screening of SCN2A in schizophrenia and identification of a novel loss-of-function mutation Psychiatr. Genet. 26 2016 60 65 10.1097/ypg.0000000000000110 26555645 PMC4756433 source

In the paper

“Transcriptomic and epigenomic studies in human tissue and AD mouse models have identified multiple cell states of microglia, representing a spectrum between homeostatic and disease-associated microglia (DAMs).12,50,51”

What the sources say

• Supported

  The Results section, “Microglia trajectory to define disease-associated microglia,” describes a trajectory with decreasing homeostatic and increasing stage 1 disease-associated signatures. It explicitly interprets that trajectory as a transition from a homeostatic to a disease-associated cell state. (LLM, conf=0.93)

  cites [12] Morabito S. Miyoshi E. Michael N. Shahin S. Martini A.C. Head E. Silva J. Leavy K. Perez-Rosendahl M. Swarup V. Single-nucleus chromatin accessibility and transcriptomic characterization of Alzheimer’s disease Nat. Genet. 53 2021 1143 1155 10.1038/s41588-021-00894-z 34239132 PMC8766217 source
• Supported

  The Results describe homeostatic microglia and disease-associated microglia (DAM) at opposite ends of a graph, with intermediate cells connecting them. The paper also reports a transition between these states as disease progresses in an AD mouse model. (LLM, conf=0.93)

  cites [50] Keren-Shaul H. Spinrad A. Weiner A. Matcovitch-Natan O. Dvir-Szternfeld R. Ulland T.K. David E. Baruch K. Lara-Astaiso D. Toth B. A unique microglia type associated with restricting development of Alzheimer’s disease Cell 169 2017 1276 1290.e17 10.1016/j.cell.2017.05.018 28602351 source
• Partly supported

  The Discussion (page not provided) describes multiple microglial states on a continuous spectrum from homeostatic cells along two response branches. It says the ARMs response overlaps with DAMs, but cautions that calling these cells “disease-associated” is misleading because they also occur in healthy mice. The paper supports the spectrum claim, but not a simple homeostatic-to-DAM continuum. (LLM, conf=0.86)

  cites [51] Sala Frigerio C. Wolfs L. Fattorelli N. Thrupp N. Voytyuk I. Schmidt I. Mancuso R. Chen W.-T. Woodbury M.E. Srivastava G. The major risk factors for Alzheimer’s disease: age, sex, and genes modulate the microglia response to A plaques Cell Rep. 27 2019 1293 1306.e6 10.1016/j.celrep.2019.03.099 31018141 PMC7340153 sourceExperimental — accuracy not yet measured

In the paper

“CD163, a hub gene of MG-M4, is known to be involved in the breakdown of the blood-brain barrier.53,54”

What the sources say

• Partly supported

  The section “Co-Localization of CD163 Expression and Its Ligand Hp-Hb Complex” reports that blood-brain barrier breakdown was associated with microglial CD163 expression in encephalitic macaque brain. It proposes that barrier damage leads to CD163 up-regulation, not that CD163 causes or contributes to the breakdown; the claim overstates that relationship. (LLM, conf=0.86)

  cites [53] Borda J.T. Alvarez X. Mohan M. Hasegawa A. Bernardino A. Jean S. Aye P. Lackner A.A. CD163, a marker of perivascular macrophages, is up-regulated by microglia in simian immunodeficiency virus encephalitis after haptoglobin-hemoglobin complex stimulation and is suggestive of breakdown of the blood-brain barrier Am. J. Pathol. 172 2008 725 737 10.2353/ajpath.2008.070848 18276779 PMC2258269 sourceExperimental — accuracy not yet measured
• Partly supported

  The Discussion (page not supplied) reports that some CD163-positive microglia coincided with fibrinogen leakage, an indication of blood–brain barrier breakdown. It suggests that vascular compromise might allow CD163-positive macrophages to enter the brain, but does not establish that CD163 itself contributes to the breakdown or that such a role is known. (LLM, conf=0.86)

  cites [54] Pey P. Pearce R.K.B. Kalaitzakis M.E. Griffin W.S.T. Gentleman S.M. Phenotypic profile of alternative activation marker CD163 is different in Alzheimer’s and Parkinson’s disease Acta Neuropathol. Commun. 2 2014 21 10.1186/2051-5960-2-21 24528486 PMC3940003 sourceExperimental — accuracy not yet measured

In the paper

“Our previous study defined a set of transcription factors, genes, and cis-regulatory elements involved in the shift between homeostatic and DAM cell states in human AD, identifying shared and distinct signatures compared with the DAM signature from 5xFAD mice.12”

What the sources say

• Supported

  Split into 2 atoms — 2 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  cites [12] Morabito S. Miyoshi E. Michael N. Shahin S. Martini A.C. Head E. Silva J. Leavy K. Perez-Rosendahl M. Swarup V. Single-nucleus chromatin accessibility and transcriptomic characterization of Alzheimer’s disease Nat. Genet. 53 2021 1143 1155 10.1038/s41588-021-00894-z 34239132 PMC8766217 source

In the paper

“Co-expression networks have been successful for analyzing bulk proteomics datasets in human disease samples,67,68”

What the sources say

• Supported

  The Results section reports using co-expression network analysis on proteomic data from five human brain-bank cohorts and identifying modules associated with Alzheimer’s disease. The Discussion describes these as robust disease-associated proteomic networks, supporting the broad claim that the approach has been successful for human disease proteomics datasets. (LLM, conf=0.92)

  cites [67] Swarup V. Chang T.S. Duong D.M. Dammer E.B. Dai J. Lah J.J. Johnson E.C.B. Seyfried N.T. Levey A.I. Geschwind D.H. Identification of conserved proteomic networks in neurodegenerative dementia Cell Rep. 31 2020 107807 10.1016/j.celrep.2020.107807 PMC8221021 32579933 source
• Supported

  The Results section, “Construction and Validation of a Consensus AD Protein Co-Expression Network” (page not provided), reports a robust network built from mass spectrometry proteomics of more than 450 human brain tissues. Its modules were preserved in an independent cohort, and many correlated with Alzheimer’s disease pathology and cognitive function. This supports the broad claim that co-expression networks have been successful in analyzing bulk proteomics data from human disease samples. (LLM, conf=0.93)

  cites [68] Johnson E.C.B. Dammer E.B. Duong D.M. Ping L. Zhou M. Yin L. Higginbotham L.A. Guajardo A. White B. Troncoso J.C. Large-scale proteomic analysis of Alzheimer’s disease brain and cerebrospinal fluid reveals early changes in energy metabolism associated with microglia and astrocyte activation Nat. Med. 26 2020 769 780 10.1038/s41591-020-0815-6 32284590 PMC7405761 source

In the paper

“These studies both used AD samples and controls from the same patient cohorts (Religious Orders Study and Memory and Aging Project, Mayo Clinic, Mount Sinai School of Medicine),64,65,66 but they took unique approaches for co-expression network analysis.”

What the sources say

• Partly supported

  The Results section, “Data origin and phenotypes” (page unavailable), identifies this study’s participants as coming from the Religious Orders Study and Rush Memory and Aging Project; the RNA-Seq results describe analysis of 478 participants. The Introduction and “Constructing the nodes of our module-trait network” describe its co-expression network approach and how it differs from prior approaches. Our copy does not establish that both studies used the same cohorts, including Mayo Clinic and Mount Sinai, so that comparison is not supported by this source. (LLM, conf=0.86)

  cites [64] Mostafavi S. Gaiteri C. Sullivan S.E. White C.C. Tasaki S. Xu J. Taga M. Klein H.-U. Patrick E. Komashko V. A molecular network of the aging human brain provides insights into the pathology and cognitive decline of Alzheimer’s disease Nat. Neurosci. 21 2018 811 819 10.1038/s41593-018-0154-9 29802388 PMC6599633 sourceExperimental — accuracy not yet measured
• Supported

  The citation follows the cohort list; this paper supports the Mayo Clinic portion of that list. Methods > Study Populations reports RNA-seq samples from people with AD and controls in the Mayo cohort. The sentence’s comparison of co-expression approaches is outside the portion supported by this citation. (LLM, conf=0.89)

  cites [65] Allen M. Carrasquillo M.M. Funk C. Heavner B.D. Zou F. Younkin C.S. Burgess J.D. Chai H.-S. Crook J. Eddy J.A. Human whole genome genotype and transcriptome data for Alzheimer’s and other neurodegenerative diseases Sci. Data 3 2016 160089 10.1038/sdata.2016.89 PMC5058336 27727239 source
• Supported

  The citation is attached to the list of patient cohorts, including Mount Sinai, rather than to the comparison of co-expression methods. The Background & Summary identifies this source as RNA-seq data from postmortem control and AD brains, and Methods identifies the Mount Sinai brain-bank cohort. No page numbers are provided in our copy. (LLM, conf=0.90)

  cites [66] Wang M. Beckmann N.D. Roussos P. Wang E. Zhou X. Wang Q. Ming C. Neff R. Ma W. Fullard J.F. The Mount Sinai cohort of large-scale genomic, transcriptomic and proteomic data in Alzheimer’s disease Sci. Data 5 2018 180185 10.1038/sdata.2018.185 PMC6132187 30204156 source

In the paper

“and in another study we showed that scWGCNA was suitable for network analysis using scRNA-seq alone.28”

What the sources say

• Supported

  The STAR★METHODS Quantification and Statistical Analysis section reports that the authors performed scWGCNA on gene-expression profiles from individual nuclei, constructed co-expression networks, and identified modules. The Method Details section identifies the input as single-nucleus RNA-sequencing data. This demonstrates network analysis using sequencing-derived expression data alone, although the experiment used nuclei rather than whole cells; page numbers were not available in our copy. (LLM, conf=0.86)

  cites [28] Kiani Shabestari S. Morabito S. Danhash E.P. McQuade A. Sanchez J.R. Miyoshi E. Chadarevian J.P. Claes C. Coburn M.A. Hasselmann J. Absence of microglia promotes diverse pathologies and early lethality in Alzheimer’s disease mice Cell Rep. 39 2022 110961 10.1016/j.celrep.2022.110961 PMC9285116 35705056 source

In the paper

“Contrasting the hdWGCNA package, the implementation of scWGCNA was an R package containing a single function for metacell construction, and a single tutorial to cover the basics of network analysis using the WGCNA package1 with the metacell matrix.”

What the sources say

• Supported

  The citation is attached to “the WGCNA package,” not to the description of scWGCNA’s implementation. The Results section (“Overview of functions included in the WGCNA package”) identifies WGCNA as an R package for correlation network analysis, which supports that identification. (LLM, conf=0.91)

  cites [1] Langfelder P. Horvath S. WGCNA: an R package for weighted correlation network analysis BMC Bioinf. 9 2008 559 10.1186/1471-2105-9-559 PMC2631488 19114008 source

In the paper

“Unlike other network analysis pipelines such as single-cell regulatory network inference and clustering (SCENIC)70 or CellChat,71 hdWGCNA is a purely unsupervised approach and does not require prior knowledge or databases in the inference procedure.”

What the sources say

• Supported

  The citation to SCENIC supports the contrast with hdWGCNA, not the claim about hdWGCNA itself. SCENIC’s Online Methods (RcisTarget; page unavailable) state that its inference workflow uses a database of motif rankings and identify the motif databases used in the paper. (LLM, conf=0.92)

  cites [70] Aibar S. González-Blas C.B. Moerman T. Huynh-Thu V.A. Imrichova H. Hulselmans G. Rambow F. Marine J.-C. Geurts P. Aerts J. SCENIC: single-cell regulatory network inference and clustering Nat. Methods 14 2017 1083 1086 10.1038/nmeth.4463 28991892 PMC5937676 source
• Supported

  The citation is attached to CellChat as a contrasting example, not to the claim about hdWGCNA itself. The Results (Overview of CellChat; page unavailable) describe CellChatDB as a literature-supported database of known ligand–receptor interactions used to predict intercellular communication. This supports the attributed contrast: CellChat’s inference uses prior biological knowledge rather than being purely unsupervised. (LLM, conf=0.91)

  cites [71] Jin S. Guerrero-Juarez C.F. Zhang L. Chang I. Ramos R. Kuan C.-H. Myung P. Plikus M.V. Nie Q. Inference and analysis of cell-cell communication using CellChat Nat. Commun. 12 2021 1088 10.1038/s41467-021-21246-9 33597522 PMC7889871 source

In the paper

“The AMP-AD study from Wan et al.63 performed network analysis separately from each brain region, while, in our previous study,56 we performed consensus network analysis across the different brain regions.”

What the sources say

• Supported

  The STAR Methods (Network Inference and Module Identification) state that network module identification methods were applied to each of seven tissue-specific datasets. The Results (A Consensus Atlas of AD-Associated Transcriptional Modules from Human Brain) report brain region-specific coexpression and consensus modules. The paper also compares modules across regions later, but that does not change how the initial network analyses were performed. (LLM, conf=0.94)

  cites [63] Wan Y.-W. Al-Ouran R. Mangleburg C.G. Perumal T.M. Lee T.V. Allison K. Swarup V. Funk C.C. Gaiteri C. Allen M. Wang M. Meta-analysis of the Alzheimer’s disease human brain transcriptome and functional dissection in mouse models Cell Rep. 32 2020 107908 10.1016/j.celrep.2020.107908 PMC7428328 32668255 source

In the paper

“hdWGCNA greatly extends upon scWGCNA,12 our previous method for co-expression network analysis in single-cell transcriptomics data.”

What the sources say

• Supported

  The Methods section says the authors developed scWGCNA as a co-expression network analysis approach for single-cell data. The Discussion likewise describes it as a method they introduced for cell population-specific co-expression networks. The citation identifies the earlier method; the claim that hdWGCNA extends it is the citing authors’ own comparison. (LLM, conf=0.93)

  cites [12] Morabito S. Miyoshi E. Michael N. Shahin S. Martini A.C. Head E. Silva J. Leavy K. Perez-Rosendahl M. Swarup V. Single-nucleus chromatin accessibility and transcriptomic characterization of Alzheimer’s disease Nat. Genet. 53 2021 1143 1155 10.1038/s41588-021-00894-z 34239132 PMC8766217 source

In the paper

“The AMP-AD study from Wan et al.63 performed network analysis separately from each brain region, while, in our previous study,56 we performed consensus network analysis across the different brain regions.”

What the sources say

• Supported

  The Results (Figure 1) describe consensus weighted gene co-expression network analysis using datasets from several brain regions. The Materials and Methods section, “Consensus WGCNA,” confirms that the study performed this analysis across the ROSMAP, Mayo, and MSSM datasets. (LLM, conf=0.94)

  cites [56] Morabito S. Miyoshi E. Michael N. Swarup V. Integrative genomics approach identifies conserved transcriptomic networks in Alzheimer’s disease Hum. Mol. Genet. 29 2020 2899 2919 10.1093/hmg/ddaa182 32803238 PMC7566321 source

In the paper

“scWGCNA was originally used to identify co-expression networks using bulk and single-cell RNA-seq together,12”

What the sources say

• Supported

  The Methods section says the authors developed scWGCNA by integrating single-nucleus and bulk-tissue RNA-seq datasets. The Results section says they used it to jointly form co-expression networks from those datasets. Single-nucleus RNA-seq is the specific form of single-cell data used here. (LLM, conf=0.94)

  cites [12] Morabito S. Miyoshi E. Michael N. Shahin S. Martini A.C. Head E. Silva J. Leavy K. Perez-Rosendahl M. Swarup V. Single-nucleus chromatin accessibility and transcriptomic characterization of Alzheimer’s disease Nat. Genet. 53 2021 1143 1155 10.1038/s41588-021-00894-z 34239132 PMC8766217 source

In the paper

“Connected genes in biological networks are potentially involved in the same processes, and EGAD evaluates this network property given a set of gene-process annotations.”

What the sources say

• Supported

  The “Neighbor-voting for network analysis and gene function prediction” section states the guilt-by-association principle that genes with shared functions are preferentially connected. It also says EGAD takes a gene network and gene annotations as input and measures how well connected genes align with an annotated functional group. (LLM, conf=0.94)

  cites [23] Ballouz S. Weber M. Pavlidis P. Gillis J. EGAD: ultra-fast functional analysis of gene networks Bioinformatics 33 2017 612 614 10.1093/bioinformatics/btw695 27993773 PMC6041978 source

In the paper

“Connected genes in biological networks are potentially involved in the same processes, and EGAD evaluates this network property given a set of gene-process annotations.”

What the sources say

• Supported

  The Abstract says EGAD assesses how well a gene network groups known sets of genes. The “Gene networks and gene annotation sets” section defines those sets as functional groups, including biological processes, represented by gene–group annotations. Together, these passages support the description of EGAD as evaluating functional coherence using gene-process annotations; page numbers were not provided. (LLM, conf=0.92)

  cites [23] Ballouz S. Weber M. Pavlidis P. Gillis J. EGAD: ultra-fast functional analysis of gene networks Bioinformatics 33 2017 612 614 10.1093/bioinformatics/btw695 27993773 PMC6041978 source

In the paper

“among which a gene set by Jansen et al.46 provided the set of genes associated with AD.”

What the sources say

• Supported

  The Results section (‘Implicated genes’) reports 192 uniquely mapped genes linked to AD-associated variants. The Results section (‘Gene-sets implicated in AD and AD-by-proxy’) also identifies gene-sets significantly associated with AD risk. These passages support citing Jansen et al. as a source of AD-associated genes. (LLM, conf=0.90)

  cites Jansen et al.46 Jansen I.E. Savage J.E. Watanabe K. Bryois J. Williams D.M. Steinberg S. Sealock J. Karlsson I.K. Hägg S. Athanasiu L. Genome-wide meta-analysis identifies new loci and functional pathways influencing Alzheimer’s disease risk Nat. Genet. 51 2019 404 413 10.1038/s41588-018-0311-9 30617256 PMC6836675 source

In the paper

“Notably, SEACells and Metacell2 do not account for cell labels in their aggregation procedures, which may result in a number of metacells containing transcriptomes from differently labeled cells.”

What the sources say

• Partly supported

  The Methods sections on graph construction and the divide-and-conquer algorithm describe forming Metacell-2 metacells from gene-expression correlations and graph partitions, without using cell labels. That supports the Metacell-2 portion and makes mixed-label metacells a plausible possibility, but the paper does not establish how many such metacells occur or address SEACells. (LLM, conf=0.83)

  cites [14] Ben-Kiki O. Bercovich A. Lifshitz A. Tanay A. Metacell-2: a divide-and-conquer metacell algorithm for scalable scRNA-seq analysis Genome Biol. 23 2022 100 10.1186/s13059-022-02667-1 35440087 PMC9019975 sourceExperimental — accuracy not yet measured
• Supported

  The Methods (SEACells algorithm; no page given) describe assigning cells to metacells by archetypal analysis of a cell-similarity matrix, then annotating the resulting metacells by their most frequent cell type. The Methods (Robustness of SEACells algorithm; no page given) assess cell-type purity for both SEACells and MetaCell-2 metacells, consistent with metacells being able to contain differently labeled cells. (LLM, conf=0.86)

  cites [16] Persad S. Choo Z.-N. Dien C. Sohail N. Masilionis I. Chaligné R. Nawy T. Brown C.C. Sharma R. Pe’er I. SEACells infers transcriptional and epigenomic cellular states from single-cell genomics data Nat. Biotechnol. 2023 10.1038/s41587-023-01716-9 PMC10713451 36973557 source

In the paper

“Of the 121,451 cells in this dataset, 20,249 were labeled as inhibitory neurons based on marker gene expression profiles.”

What the sources say

• Partly supported

  The paper’s clustering passage (section labeled Abstract; page unavailable) says it annotated cell types, including interneurons, using known cell-type markers. Its sequencing passage reports 104,559 ASD-and-control nuclei plus 21,984 additional epilepsy-and-control profiles; our copy does not establish the claimed total of 121,451 cells or the count of 20,249 inhibitory neurons. (LLM, conf=0.85)

  cites Velmeshev et al.9 Velmeshev D. Schirmer L. Jung D. Haeussler M. Perez Y. Mayer S. Bhaduri A. Goyal N. Rowitch D.H. Kriegstein A.R. Single-cell genomics identifies cell type–specific molecular changes in autism Science 364 2019 685 689 10.1126/science.aav8130 31097668 PMC7678724 sourceExperimental — accuracy not yet measured

In the paper

“Beyond conventional single-cell RNA-seq, hdWGCNA is capable of performing isoform-level network analysis using long-read single-cell data.”

In the paper

“hdWGCNA is directly compatible with Seurat, a widely used R package for single-cell and spatial transcriptomics analysis”

In the paper

“hdWGCNA provides functions for network inference, gene module identification, gene enrichment analysis, statistical tests, and data visualization.”

In the paper

“hdWGCNA computes MEs using specific accommodations for high-dimensional data, allowing for batch correction and regression of continuous covariates (STAR Methods, Algorithm 2).”

In the paper

“The runtime of the MetacellsByGroups function increased steadily with the number of cells, but the memory usage plateaus.”

In the paper

“we used hdWGCNA to analyze a single-cell RNA sequencing (scRNA-seq) dataset consisting of 1 million cells, showcasing the scalability of hdWGCNA in large datasets.”

In the paper

“The inhibitory neuron network performed the best for functional coherence with a median neighbor-voting AUC of 0.592, while the lowest-performing network was from the oligodendrocytes with a median AUC of 0.549.”

In the paper

“we designed hdWGCNA to facilitate multi-scale analysis of cellular and spatial hierarchies.”

In the paper

“While TestSoftPowers generally had a low memory footprint, it was the slowest individual function based on these tests.”

In the paper

“The efficiency of ConstructNetwork varies both with the number of input cells and features, where this calculation will slow down as more cells and features are included.”

In the paper

“Finally, ModuleConnectivity computes eigengene-based connectivity as product-moment correlation coefficients between the sparse gene expression matrix and the MEs matrix, which resulted in fast calculations with low memory usage.”

In the paper

“we used hdWGCNA to project gene modules from two bulk RNA-seq studies of AD patients into an snRNA-seq dataset of the AD brain, showing that our approach allows for interrogation of gene modules and networks that have been previously identified.”

In the paper

“we performed a similar iterative network analysis on a peripheral blood mononuclear cell (PBMC) scRNA-seq dataset of nearly 1M cells, highlighting the scalability of hdWGCNA in large datasets (Figure S5; Data S1).”

In the paper

“Further, the efficiency of this function varies with the number of co-expression modules detected and with the number of features in each modules.”

In the paper

“we performed a pairwise gene set overlap analysis of the 96 co-expression modules, and, while we did find that some modules had significant overlaps across the different cell types, the gene sets comprising these modules were overall quite distinct, with a maximum Jaccard index between two modules of 0.297 and a median of 0.005”

In the paper

“Overall, we found that module hub gene expression was generally predictive of module member gene expression across all modules in the six cell-type co-expression networks, where the module with the best performance had an average test set RMSE of 0.0159 and the module with the worst performance had an average test set RMSE of 0.209 (Figure 2E).”

In the paper

“hdWGCNA provides a rich suite of functions for data analysis and visualization, providing biological context for co-expression networks by leveraging a variety of biological knowledge databases.”

In the paper

“hdWGCNA is capable of processing any number of ST samples in the same co-expression network analysis by constructing metaspots separately for each sample.”

In the paper

“Further, the hub genes of SM4 include hemoglobin subunits (Hba-a1, Hba-a2, Hbb-bt), and we show that SM4 was enriched for biological processes associated with brain vasculature (Figures 3B and S7C).”

In the paper

“Different isoforms of the histone H3.3 subunit gene H3f3b were hubs for modules RGL-M1 and RGL-M11, which were associated with ependymal and neuronal cell fates respectively, suggesting that alternative TES usage in H3f3b plays a role in regulating epigenetic factors in murine hippocampal development (Figure 4F).”

In the paper

“the module eigengene algorithm in hdWGCNA (Algorithm 2) accounts for batch effects and additional covariates in the input dataset”

In the paper

“We show that expression of these modules was significantly correlated with AD genetic risk (Pearson correlation p <0.05), with the strongest correlation in alternative activation module MG-M4 (Figure 6I).”

In the paper

“hdWGNCA is completely technology agnostic, and can be adapted to handle high dimensional counts matrices from any single-cell or spatial transcriptomics platform.”

In the paper

“and hdWGCNA contains functions to handle spatial transcriptomics datasets.”

In the paper

“In this study, we demonstrated that hdWGCNA is compatible with single-cell and ST datasets and can be easily adapted for novel transcriptomics approaches such as ScISOrSeq.”

In the paper

“Unlike scWGCNA, the metacell construction algorithm (Algorithm 1) in hdWGCNA includes new parameters to avoid redundant metacells”

In the paper

“The individually processed datasets were merged into a single Seurat object comprising 189,127 nuclei”

In the paper

“The major cell compartments recovered in this analysis were similar to those reported by Parse Biosciences in their analysis, including as T-cells, B-cells, monocytes, dendritic cells, basophils, and plasmablasts.”

In the paper

“genes expressed in at least 5% of microglia from any of the three studies were retained (7,900 genes).”

In the paper

“The metacell aggregation step had a runtime of 85 min and 59 s.”

In the paper

“We retained all nuclei labeled microglia for network analysis based on expression of canonical marker genes such as CSF1R (9,904 nuclei)”

In the paper

“As shown throughout this manuscript, hdWGCNA includes functions for projecting co-expression networks into a variety of external datasets.”

In the paper

“While ST spots were clustered based on transcriptomic information alone, we were able to annotate them based on anatomical features.”

In the paper

“We note that for the largest cell population (T-cells, 555,417 cells), the runtime for the network construction step was 186 s.”

In the paper

“Importantly, the number of input genes and other network analysis parameters also have an effect on runtime and memory usage.”

In the paper

“This function attempts to construct a target number of metacells within each biological replicate and each cell population, and the algorithm terminates early if this target is reached, thus explaining the plateau in the memory usage graph.”

In the paper

“Here we describe hdWGCNA, a comprehensive framework for constructing and analyzing co-expression networks in high-dimensional transcriptomic data (Figure 1A).”

In the paper

“A typical hdWGCNA workflow in scRNA-seq data accounts for these considerations by collapsing highly similar cells into “metacells” to reduce sparsity while retaining cellular heterogeneity and by allowing for a modular design to perform separate network analyses in specified cell populations.”

In the paper

“For example, the hdWGCNA metacell algorithm requires a dimensional reduction of the input expression matrix, but these reductions often contain technical artifacts. The choice of dimensionality reduction method and handling of technical artifacts would then influence the effectiveness of metacell construction.”

In the paper

“Gfap encodes a key intermediate filament protein in astrocytes that is involved in astrocytic reactivity during central nervous system (CNS) injuries or neurodegeneration,39”

What the sources say

• ?Not checked — no full text of the sourcecites [39] Hol E.M. Pekny M. Glial fibrillary acidic protein (GFAP) and the astrocyte intermediate filament system in diseases of the central nervous system Curr. Opin. Cell Biol. 32 2015 121 130 10.1016/j.ceb.2015.02.004 25726916 source

In the paper

“Further, the optimal number of cells to merge together to form a single metacell may differ across cell types and tissues, attempting to balance between increasing information content of the aggregated group while avoiding merging of dissimilar cells.”

In the paper

“We also find enrichment of several ASD-associated genes such as TSC1 (INH-M8), SMARCA4 (INH-M8), SHANK2 (INH-M4), and CPEB4 (INH-M1), highlighting that these modules are functional and provide new insights into the role of inhibitory neurons in ASD.”

In the paper

“The hdWGCNA R package directly extends the familiar Seurat pipeline and the SeuratObject data structure, enabling researchers to rapidly incorporate network analysis into their own workflows, going beyond cell clustering and differential gene expression analysis toward systems-level insights.”

In the paper

“This metacell bagging algorithm is implemented as part of the hdWGCNA R package in the ConstructMetacells function, and the pseudocode for this algorithm is defined in Algorithm 1.”

In the paper

“We designed hdWGCNA to be highly modular, allowing for multi-scale analyses of different cellular or spatial hierarchies in a technology-agnostic manner.”

In the paper

“Contrasting the hdWGCNA package, the implementation of scWGCNA was an R package containing a single function for metacell construction, and a single tutorial to cover the basics of network analysis using the WGCNA package1 with the metacell matrix.”

In the paper

“In the signed TOM, negative correlations serve to negatively reinforce the network connection, which is not the case in the unsigned TOM.”

In the paper

“The hdWGCNA R package directly extends the familiar Seurat pipeline and the SeuratObject data structure, enabling researchers to rapidly incorporate network analysis into their own workflows, going beyond cell clustering and differential gene expression analysis toward systems-level insights.”

In the paper

“By projecting co-expression modules derived from bulk RNA-seq patient cohorts into single-cell datasets, we can layer disease-related information onto the single-cell dataset and attribute cell-state-specific expression patterns to the bulk RNA-seq data.”

In the paper

“We developed hdWGCNA to provide a succinct methodology for investigating systems-level changes in the transcriptome in single-cell or ST datasets.”

In the paper

“hdWGCNA includes built-in functions to leverage external biological knowledge sources to provide insight for co-expression networks, for example by comparing gene modules with functional gene sets such as disease-associated genes from GWAS expression quantitative trait loci (eQTLs), or transcription factor target genes.”

In the paper

“The co-expression information computed by hdWGCNA can be easily retrieved from the Seurat object to facilitate custom downstream analyses beyond the hdWGCNA package.”

In the paper

“we applied Harmony22 to correct MEs based on the dataset of origin.”

In the paper

“Module eigengenes were computed for each of these bulk RNA-seq modules in the snRNA-seq dataset using the hdWGCNA function ProjectModules, using Harmony to correct MEs based on sequencing batch.”

In the paper

“the set of modules from Morabito et al. 202056 was based on a consensus network analysis across six brain regions”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites Morabito et al. 202056

In the paper

“We performed a separate parameter sweep for the three expression matrices using the hdWGCNA function TestSoftPowerConsensus, ensuring that we used an appropriate β value for each dataset (Mathys et al.: β=6, Zhou et al.: β=8, Morabito & Miyoshi et al.: β=6).”

In the paper

“the datasets were integrated into a common dimensionally-reduced space using PCA and Harmony.”

In the paper

“we applied Harmony22 to correct MEs based on sequencing batch.”

In the paper

“the other set of modules from the AMP-AD study63 were constructed separately for seven different brain regions.”

What the sources say

• ?Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  cites AMP-AD study63

In the paper

“We selected a soft-power threshold β=9 based on the parameter sweep performed with the TestSoftPowers function.”

In the paper

“We oriented the start of pseudotime based on the expression of homeostatic microglia marker genes, such as P2RY12, CX3CR1, and CSF1R.”

In the paper

“Individual TOMs were computed for each dataset, and they were scaled based on the 80th percentile in order to alleviate different statistical properties specific to each dataset rather than the underlying biology.”