[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EFKFP8GCVVGAEFFD7QCD5T":3,"skill-same:01M7EFKFP8GCVVGAEFFD7QCD5T":68},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":30,"commitSha":31,"skillDirectory":32,"forksCount":14,"repositoryHtmlUrl":33,"repositoryHomepageUrl":33,"ownerHtmlUrl":34,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":35,"install":36,"resources":49},"01M7EFKFP8GCVVGAEFFD7QCD5T","bio-single-cell-multimodal-integration",null,"Integrate multimodal single-cell data (CITE-seq RNA+protein, 10x Multiome RNA+ATAC, unpaired\u002Fdiagonal RNA+ATAC) and choose the right joint method. Use when classifying an integration task by anchor structure (paired vs unpaired), denoising CITE-seq ADT background before joint embedding, picking between WNN, totalVI, MultiVI, MOFA+, GLUE, or Seurat v5 bridge integration, or diagnosing why a modality dominates a joint clustering.","Mr-Nobody-Anonymous\u002FAll-skills","Mr-Nobody-Anonymous","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F184253119?v=4","01M7AEXFJHKTYQW5R0ZCQQTWFV","anonymous","01M7C00M3GQR4QKP9GZA42GFAK",0,"MIT","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-single-cell-multimodal-integration\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: bio-single-cell-multimodal-integration\ndescription: Integrate multimodal single-cell data (CITE-seq RNA+protein, 10x Multiome RNA+ATAC, unpaired\u002Fdiagonal RNA+ATAC) and choose the right joint method. Use when classifying an integration task by anchor structure (paired vs unpaired), denoising CITE-seq ADT background before joint embedding, picking between WNN, totalVI, MultiVI, MOFA+, GLUE, or Seurat v5 bridge integration, or diagnosing why a modality dominates a joint clustering.\ntool_type: mixed\nprimary_tool: Seurat\nsource: \"https:\u002F\u002Fgithub.com\u002FGPTomics\u002FbioSkills\"\nsource_repository: \"GPTomics\u002FbioSkills\"\nsource_path: \"single-cell\u002Fmultimodal-integration\u002FSKILL.md\"\nlicense: \"MIT\"\nimported_at: \"2026-09-20\"\n---\n\n## Version Compatibility\n\nReference examples tested with: scanpy 1.10+, Seurat 5.0+, anndata 0.10+\n\nBefore using code patterns, verify installed versions match. If versions differ:\n- Python: `pip show \u003Cpackage>` then `help(module.function)` to check signatures\n- R: `packageVersion('\u003Cpkg>')` then `?function_name` to verify parameters\n\nIf code throws ImportError, AttributeError, or TypeError, introspect the installed\npackage and adapt the example to match the actual API rather than retrying.\n\n# Multimodal Integration\n\n**\"Jointly analyze my CITE-seq \u002F Multiome \u002F unpaired multi-omic data\"** -> Classify the task by anchor structure, denoise each modality in its native pipeline, then build one joint representation.\n- R: `Seurat::FindMultiModalNeighbors()` (WNN), `Signac` (ATAC LSI), `dsb::DSBNormalizeProtein()` (ADT denoising), `PrepareBridgeReference()` (v5 bridge)\n- Python: `muon`\u002F`mudata` (MuData container), `scvi.model.TOTALVI` \u002F `MULTIVI`, `MOFA2`\u002F`muon.tl.mofa`, `scglue` (diagonal)\n\n## Governing Principle\n\nClassify the integration task by its anchor structure FIRST, because the anchor decides which algorithm class is even applicable (Argelaguet 2021).\n- Horizontal: same modality, different cells; anchor = shared features (batch correction, not this skill).\n- Vertical (paired, same cell): multiple modalities measured in the same cells; anchor = shared cells. CITE-seq, 10x Multiome.\n- Diagonal (unpaired): different modalities in different cells, no shared cells and no shared features; correspondence is inferred from prior knowledge. Independent scRNA + scATAC.\n- Mosaic: partially observed grid of (modalities x batches); some blocks present, some missing.\n\nPaired vs unpaired is the master fork: paired correspondence is known a priori (WNN, totalVI, MultiVI, MOFA+), unpaired\u002Fdiagonal correspondence must be inferred (GLUE, Seurat v5 bridge), mosaic mixes both (MultiVI, StabMap). Two separately-paired datasets that share only one modality (for example a 10x Multiome and a CITE-seq experiment sharing only RNA) are a mosaic problem: anchor on the shared RNA and impute or bridge the modality-specific blocks with StabMap, MultiVI, or Seurat v5 bridge integration rather than forcing a single WNN.\n\nCITE-seq ADT background is a three-part mixture, not one \"ambient\" term: (1) ambient antibody captured in every droplet including empties, (2) cell-intrinsic non-specific binding (Fc receptors, sticky dying cells) that does NOT appear in empties, (3) spillover\u002Findex hopping between barcodes. Denoise ADT (DSB or totalVI's built-in background mixture) BEFORE any joint embedding; raw or CLR-only ADT carries this background into the joint graph.\n\nWNN can be dominated by the noisier modality: weights reward local neighbor predictability, and a handful of high-variance or saturating ADT features can manufacture self-consistent neighborhoods and get up-weighted despite carrying less biology. Report the per-cell weight distribution and check whether clustering survives down-weighting the suspect modality.\n\nImputed modalities are inferences, not measurements: MultiVI\u002FStabMap\u002FCobolt impute the missing modality for unpaired cells, and gene-activity scores from ATAC approximate RNA. Differential expression or marker calls on imputed values are model-dependent and must be flagged as such.\n\n## Classify the Task: Anchor Structure -> Method Class\n\n| Anchor structure | What is shared | Example assay | Method class |\n|---|---|---|---|\n| Vertical \u002F paired | Same cells | CITE-seq, 10x Multiome | WNN, totalVI, MultiVI(paired), MOFA+, mojitoo |\n| Diagonal \u002F unpaired | Nothing (prior graph) | Independent scRNA + scATAC | GLUE, Seurat v5 bridge, LIGER |\n| Mosaic | Some modalities only | Batch A RNA+ATAC, batch B RNA | MultiVI, StabMap, Cobolt, totalVI(partial) |\n\nWhen methods compete, verify the current best-practice default against the installed tool docs before committing; the field moves and defaults drift across minor versions.\n\n## Method Decision Table (Paired CITE-seq \u002F Multiome)\n\n| Method | Model \u002F assumption | Use when | Fails when |\n|---|---|---|---|\n| WNN (Seurat) | Per-cell, per-modality weights from cross-modality neighbor prediction; one weighted graph | Fast joint embedding\u002Fclustering of one well-normalized paired dataset | Protein background not removed upstream; noisy\u002Fsaturating modality dominates; not for unpaired\u002Fmosaic |\n| totalVI (scvi-tools) | Conditional VAE; RNA NB\u002FZINB, each protein a 2-component NB mixture (background+foreground) | Need denoised protein, principled DE, batch integration, merging different antibody panels | Tiny datasets (VAE overfits); no GPU and very large data; protein-specific background structure not captured by one per-cell factor |\n| MultiVI (scvi-tools) | Single joint VAE over RNA+ATAC(+protein); mosaic-capable, imputes missing modality | Paired+unpaired RNA\u002FATAC mixed (mosaic); want generative DE\u002FDA | \"batch\" key is the modality indicator, not sequencing batch; imputed modalities treated as measured |\n| MOFA+ (MOFA2) | Linear Bayesian group factor analysis; sparse factors, per-modality variance explained | Interpreting shared vs modality-specific axes of variation (exploratory\u002Fexplanatory) | Used for clustering\u002Fdenoising; likelihood mismatched to data; expecting batch correction within a view |\n| mojitoo | CCA across precomputed per-modality reductions; fast, parameter-free | Quick paired joint reduction from existing PCA\u002FLSI slots | No knob to down-weight a noisy modality; bounded by input reductions; paired only |\n\n## Method Decision Table (Unpaired \u002F Diagonal \u002F Mosaic)\n\n| Method | Model \u002F assumption | Use when | Fails when |\n|---|---|---|---|\n| GLUE (scglue) | Per-modality VAEs + prior feature graph (peak-near-gene); adversarial cell alignment | Unpaired diagonal scRNA + scATAC; want regulatory inference as a byproduct | Genome-build\u002Fcoordinate mismatch yields an empty guidance graph and garbage alignment; adversarial over-mixing of distinct states |\n| Seurat v5 bridge | Multiome bridge dataset = dictionary linking query modality to reference modality | Mapping a query (scATAC) onto a reference built in another modality (scRNA) | Poor\u002Fbatch-mismatched bridge propagates error; rare query-only populations mislabeled |\n| StabMap | Mosaic topology from shared features; project all cells via shortest paths | Mosaic with informative unshared features that cannot be dropped | Unshared-feature chaining compounds error per hop |\n| Cobolt \u002F scMoMaT | Generative shared latent over joint + single-modality datasets | Mosaic where a generative latent is preferred over feature chaining | DE\u002Fmarker calls made on imputed values |\n\n## ADT Normalization: CLR vs DSB\n\n| Method | What it does | Use when | Fails when |\n|---|---|---|---|\n| CLR (centered log-ratio) | Rescales compositionally; Seurat `NormalizeData(method=\"CLR\", margin=2)` | Quick, no empty droplets available; small panels | Does NOT remove background; geometric-mean denominator distorted by saturating high-abundance ADTs |\n| DSB | Ambient correction from empty droplets + per-cell technical denoising via 2-component mixture + isotype controls | Raw\u002Funfiltered matrix available (needs empty droplets); want background removed before embedding | No empty droplets retained; protein-specific non-specific binding (one per-cell factor under\u002Fover-corrects); no clearly bimodal proteins |\n\nSeurat's CLR margin is genuinely ambiguous across versions (margin=2 = per-feature is the WNN-tutorial recommendation for large panels); verify with `?NormalizeData` on the installed version.\n\n## CITE-seq: Denoise ADT, Then Joint Embed (Seurat)\n\n**Goal:** Remove ADT background with DSB before WNN, because WNN does not denoise protein.\n\n**Approach:** Estimate ambient from empty droplets and per-cell technical noise from a mixture plus isotype controls, then feed denoised ADT into the standard PCA -> WNN flow.\n\n```r\nlibrary(dsb)\nlibrary(Seurat)\n\nraw \u003C- Read10X('raw_feature_bc_matrix\u002F')           # unfiltered: contains empty droplets\ncells \u003C- Read10X('filtered_feature_bc_matrix\u002F')    # called cells\n\nadt_cells \u003C- as.matrix(cells[['Antibody Capture']])\nadt_empty \u003C- as.matrix(raw[['Antibody Capture']][, setdiff(colnames(raw[['Antibody Capture']]), colnames(adt_cells))])\n\n# isotype.control.name.vec must name the ACTUAL isotype rows (often IgG1\u002FIgG2a\u002FMouse-IgG2b-Ctrl); the regex below misses those\n# When isotypes are absent or not matched, set use.isotype.control = FALSE (keep denoise.counts = TRUE) and pass real names explicitly\nadt_dsb \u003C- DSBNormalizeProtein(\n    cell_protein_matrix = adt_cells,\n    empty_drop_matrix = adt_empty,\n    denoise.counts = TRUE,\n    use.isotype.control = TRUE,\n    isotype.control.name.vec = grep('[Ii]sotype|IgG', rownames(adt_cells), value = TRUE)\n)\n```\n\n## CITE-seq: WNN Joint Clustering (Seurat)\n\n**Goal:** Build one weighted-NN graph from denoised RNA and ADT and cluster on it.\n\n**Approach:** Reduce each modality independently (PCA on RNA, PCA on the small ADT panel), then learn per-cell modality weights and cluster\u002Fembed on the joint graph.\n\n```r\nobj[['ADT']] \u003C- CreateAssay5Object(data = adt_dsb)        # DSB output is already normalized data\nDefaultAssay(obj) \u003C- 'RNA'\nobj \u003C- NormalizeData(obj) |> FindVariableFeatures() |> ScaleData() |> RunPCA(reduction.name = 'pca')\n\nDefaultAssay(obj) \u003C- 'ADT'\nVariableFeatures(obj) \u003C- rownames(obj[['ADT']])\nobj \u003C- ScaleData(obj) |> RunPCA(reduction.name = 'apca', npcs = min(18, nrow(obj[['ADT']]) - 1))\n\n# dims.list matched to informative dims; small ADT panels saturate by ~1:18\nobj \u003C- FindMultiModalNeighbors(obj, reduction.list = list('pca', 'apca'), dims.list = list(1:30, 1:18))\nobj \u003C- FindClusters(obj, graph.name = 'wsnn', algorithm = 3)   # algorithm 3 = SLM (the tutorial choice), NOT Leiden\nobj \u003C- RunUMAP(obj, nn.name = 'weighted.nn', reduction.name = 'wnn.umap')\n\n# Inspect the per-cell weight distribution; a single dominant modality is a red flag\nVlnPlot(obj, features = 'RNA.weight', group.by = 'seurat_clusters')\n```\n\n## CITE-seq: totalVI (Python, denoise + DE in one model)\n\n**Goal:** Jointly model RNA + protein with explicit protein background, yielding a denoised latent space and foreground probabilities.\n\n**Approach:** Register a MuData object, train the conditional VAE, then read the latent representation and per-protein foreground probability.\n\n```python\nimport scvi\nimport mudata as md\n\n# mdata holds .mod['rna'] (raw counts) and .mod['prot'] (raw ADT counts)\nscvi.model.TOTALVI.setup_mudata(\n    mdata, rna_layer='counts', protein_layer=None,\n    modalities={'rna_layer': 'rna', 'protein_layer': 'prot'}\n)\nmodel = scvi.model.TOTALVI(mdata)\nmodel.train()\n\nmdata.obsm['X_totalVI'] = model.get_latent_representation()\nfg = model.get_protein_foreground_probability()        # 1 - background mixing weight per protein per cell\ndenoised_rna, denoised_prot = model.get_normalized_expression()\n```\n\n## Multiome (RNA + ATAC, same cell): Native Pipelines, Then Join\n\n**Goal:** Process each modality in its own statistics before joining, because RNA and ATAC have incompatible distributions.\n\n**Approach:** PCA on RNA, TF-IDF + LSI on ATAC (drop depth-correlated components), then WNN. See scatac-analysis for ATAC QC and the binarization\u002Fdepth-component caveats.\n\n```r\nlibrary(Signac)\nDefaultAssay(obj) \u003C- 'RNA'\nobj \u003C- NormalizeData(obj) |> FindVariableFeatures() |> ScaleData() |> RunPCA()\n\nDefaultAssay(obj) \u003C- 'ATAC'\nobj \u003C- RunTFIDF(obj) |> FindTopFeatures(min.cutoff = 'q0') |> RunSVD()\nDepthCor(obj)                                          # diagnose which LSI components track depth\n\n# dims = 2:30 drops LSI_1 ONLY if DepthCor confirms it tracks depth (usually true, not guaranteed)\nobj \u003C- FindMultiModalNeighbors(obj, reduction.list = list('pca', 'lsi'), dims.list = list(1:30, 2:30))\nobj \u003C- RunUMAP(obj, nn.name = 'weighted.nn', reduction.name = 'wnn.umap')\nobj \u003C- FindClusters(obj, graph.name = 'wsnn', algorithm = 3)\n```\n\nMerging multiome datasets requires a common peak set: re-quantify all cells against unified peaks, or peak-boundary differences manufacture spurious batch structure. The ATAC gene-activity matrix is an approximation, not measured RNA; do not conflate it with the RNA modality.\n\n## MOFA+ (interpretable shared\u002Fspecific factors)\n\n**Goal:** Decompose modalities into shared latent factors with per-modality variance explained.\n\n**Approach:** Build a MOFA object from per-modality matrices, set likelihoods to match each data type, run, then interpret factor loadings.\n\n```python\nimport muon as mu\n\n# likelihoods must match data: gaussian for scaled RNA, bernoulli for binarized ATAC, poisson for counts\nmu.tl.mofa(mdata, n_factors=15, outfile='mofa_model.hdf5')   # writes mdata.obsm['X_mofa']\n```\n\n## Unpaired \u002F Diagonal: GLUE (Python)\n\n**Goal:** Align independent scRNA and scATAC with no shared cells via a prior feature graph.\n\n**Approach:** Configure each dataset with a count-appropriate probabilistic model, build a gene-anchored guidance graph, fit GLUE, then read aligned embeddings.\n\n```python\nimport scglue\n\nscglue.models.configure_dataset(rna, 'NB', use_highly_variable=True, use_rep='X_pca')     # NB needs RAW counts\nscglue.models.configure_dataset(atac, 'ZINB', use_highly_variable=True, use_rep='X_lsi')\ngraph = scglue.genomics.rna_anchored_guidance_graph(rna, atac)     # peak-near-gene prior; coords must share genome build\nglue = scglue.models.fit_SCGLUE({'rna': rna, 'atac': atac}, graph)\nrna.obsm['X_glue'] = glue.encode_data('rna', rna)\natac.obsm['X_glue'] = glue.encode_data('atac', atac)\n```\n\nVerify cell-type structure is preserved (not just modality overlap); adversarial alignment can over-mix distinct populations.\n\n## MuData Housekeeping\n\nAfter per-modality QC, modalities hold different cell sets; `muon.pp.intersect_obs(mdata)` before any paired analysis. Editing a modality-local `mdata.mod['rna'].obs` needs `mdata.update()` to propagate to the global `mdata.obs`. R round-trips (MuDataSeurat, zellkonverter) are lossy; plan to stay in one ecosystem.\n\n## Common Errors\n\n| Symptom | Cause | Fix |\n|---|---|---|\n| WNN clustering driven entirely by ADT | A few saturating high-variance proteins dominate the neighbor graph | Report per-cell weight distribution; down-weight or denoise ADT (DSB); re-check clustering stability |\n| \"Background\" smear in every ADT cluster | Ran WNN\u002FCLR without empty-droplet denoising | Run DSB (needs raw\u002Funfiltered matrix) or totalVI before joint embedding |\n| DSB errors \u002F nonsense output | Passed a filtered cell matrix only (no empty droplets) | Supply `empty_drop_matrix` from the raw\u002Funfiltered matrix |\n| Spurious batch structure after merging multiome | Per-dataset peak sets, not a unified set | Re-quantify all cells against one common peak set |\n| GLUE produces a blob \u002F no alignment | Guidance graph near-empty from genome-build\u002Fcoordinate mismatch | Align RNA gene coords and ATAC peaks to the same build before building the graph |\n| RNA and protein disagree for a marker | Often real post-transcriptional biology (stability, trafficking, lag), not an artifact | Do not \"correct away\"; treat single-gene discordance as informative |\n| MultiVI batch effects persist | The `batch_key` was set to the modality indicator, not sequencing batch | Add a separate covariate for the real batch |\n| DE on a modality looks too clean | Computed on imputed\u002Fgene-activity values, not measurements | Flag imputed-modality DE as model-dependent; validate against a measured modality |\n\n## Related Skills\n\n- single-cell\u002Fscatac-analysis - ATAC QC, TF-IDF\u002FLSI, gene-activity caveats for the Multiome ATAC half\n- single-cell\u002Fpreprocessing - per-modality RNA QC and normalization before integration\n- single-cell\u002Fclustering - clustering and UMAP on the joint graph\n- single-cell\u002Fbatch-integration - horizontal (same-modality, cross-sample) correction\n- single-cell\u002Fmarkers-annotation - marker-based interpretation of joint clusters\n- atac-seq\u002Fmotif-deviation - chromVAR TF activity on the Multiome ATAC modality\n- pathway-analysis\u002Fgo-enrichment - functional interpretation of modality-specific factors\n\n## References\n\nArgelaguet R, Cuomo ASE, Stegle O, Marioni JC. Computational principles and challenges in single-cell data integration. Nat Biotechnol 39(10):1202-1215 (2021).\nStoeckius M, Hafemeister C, Stephenson W, et al. Simultaneous epitope and transcriptome measurement in single cells (CITE-seq). Nat Methods 14:865-868 (2017).\nMulè MP, Martins AJ, Tsang JS. Normalizing and denoising protein expression data from droplet-based single-cell profiling (DSB). Nat Commun 13:2099 (2022).\nHao Y, Hao S, Andersen-Nissen E, et al. Integrated analysis of multimodal single-cell data (WNN). Cell 184(13):3573-3587 (2021).\nGayoso A, Steier Z, Lopez R, et al. Joint probabilistic modeling of single-cell multi-omic data with totalVI. Nat Methods 18:272-282 (2021).\nAshuach T, Gabitto MI, Koodli RV, et al. MultiVI: deep generative model for the integration of multimodal data. Nat Methods 20(8):1222-1231 (2023).\nArgelaguet R, Arnol D, Bredikhin D, et al. MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data. Genome Biol 21:111 (2020).\nCao Z-J, Gao G. Multi-omics single-cell data integration and regulatory inference with graph-linked unified embedding (GLUE). Nat Biotechnol 40(10):1458-1466 (2022).\nHao Y, Stuart T, Kowalski MH, et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis (Seurat v5 bridge). Nat Biotechnol 42:293-304 (2024).\nBredikhin D, Kats I, Stegle O. MUON: multimodal omics analysis framework. Genome Biol 23:42 (2022).\nGhazanfar S, Guibentif C, Marioni JC. Stabilized mosaic single-cell data integration using unshared features (StabMap). Nat Biotechnol 42(2):284-292 (2024).\nYin Y, et al. Characterization and decontamination of background noise in droplet-based single-cell protein expression data with DecontPro. Nucleic Acids Res 52(1):e4 (2024).\n","## Version Compatibility\n\nReference examples tested with: scanpy 1.10+, Seurat 5.0+, anndata 0.10+\n\nBefore using code patterns, verify installed versions match. If versions differ:\n- Python: `pip show \u003Cpackage>` then `help(module.function)` to check signatures\n- R: `packageVersion('\u003Cpkg>')` then `?function_name` to verify parameters\n\nIf code throws ImportError, AttributeError, or TypeError, introspect the installed\npackage and adapt the example to match the actual API rather than retrying.\n\n# Multimodal Integration\n\n**\"Jointly analyze my CITE-seq \u002F Multiome \u002F unpaired multi-omic data\"** -> Classify the task by anchor structure, denoise each modality in its native pipeline, then build one joint representation.\n- R: `Seurat::FindMultiModalNeighbors()` (WNN), `Signac` (ATAC LSI), `dsb::DSBNormalizeProtein()` (ADT denoising), `PrepareBridgeReference()` (v5 bridge)\n- Python: `muon`\u002F`mudata` (MuData container), `scvi.model.TOTALVI` \u002F `MULTIVI`, `MOFA2`\u002F`muon.tl.mofa`, `scglue` (diagonal)\n\n## Governing Principle\n\nClassify the integration task by its anchor structure FIRST, because the anchor decides which algorithm class is even applicable (Argelaguet 2021).\n- Horizontal: same modality, different cells; anchor = shared features (batch correction, not this skill).\n- Vertical (paired, same cell): multiple modalities measured in the same cells; anchor = shared cells. CITE-seq, 10x Multiome.\n- Diagonal (unpaired): different modalities in different cells, no shared cells and no shared features; correspondence is inferred from prior knowledge. Independent scRNA + scATAC.\n- Mosaic: partially observed grid of (modalities x batches); some blocks present, some missing.\n\nPaired vs unpaired is the master fork: paired correspondence is known a priori (WNN, totalVI, MultiVI, MOFA+), unpaired\u002Fdiagonal correspondence must be inferred (GLUE, Seurat v5 bridge), mosaic mixes both (MultiVI, StabMap). Two separately-paired datasets that share only one modality (for example a 10x Multiome and a CITE-seq experiment sharing only RNA) are a mosaic problem: anchor on the shared RNA and impute or bridge the modality-specific blocks with StabMap, MultiVI, or Seurat v5 bridge integration rather than forcing a single WNN.\n\nCITE-seq ADT background is a three-part mixture, not one \"ambient\" term: (1) ambient antibody captured in every droplet including empties, (2) cell-intrinsic non-specific binding (Fc receptors, sticky dying cells) that does NOT appear in empties, (3) spillover\u002Findex hopping between barcodes. Denoise ADT (DSB or totalVI's built-in background mixture) BEFORE any joint embedding; raw or CLR-only ADT carries this background into the joint graph.\n\nWNN can be dominated by the noisier modality: weights reward local neighbor predictability, and a handful of high-variance or saturating ADT features can manufacture self-consistent neighborhoods and get up-weighted despite carrying less biology. Report the per-cell weight distribution and check whether clustering survives down-weighting the suspect modality.\n\nImputed modalities are inferences, not measurements: MultiVI\u002FStabMap\u002FCobolt impute the missing modality for unpaired cells, and gene-activity scores from ATAC approximate RNA. Differential expression or marker calls on imputed values are model-dependent and must be flagged as such.\n\n## Classify the Task: Anchor Structure -> Method Class\n\n| Anchor structure | What is shared | Example assay | Method class |\n|---|---|---|---|\n| Vertical \u002F paired | Same cells | CITE-seq, 10x Multiome | WNN, totalVI, MultiVI(paired), MOFA+, mojitoo |\n| Diagonal \u002F unpaired | Nothing (prior graph) | Independent scRNA + scATAC | GLUE, Seurat v5 bridge, LIGER |\n| Mosaic | Some modalities only | Batch A RNA+ATAC, batch B RNA | MultiVI, StabMap, Cobolt, totalVI(partial) |\n\nWhen methods compete, verify the current best-practice default against the installed tool docs before committing; the field moves and defaults drift across minor versions.\n\n## Method Decision Table (Paired CITE-seq \u002F Multiome)\n\n| Method | Model \u002F assumption | Use when | Fails when |\n|---|---|---|---|\n| WNN (Seurat) | Per-cell, per-modality weights from cross-modality neighbor prediction; one weighted graph | Fast joint embedding\u002Fclustering of one well-normalized paired dataset | Protein background not removed upstream; noisy\u002Fsaturating modality dominates; not for unpaired\u002Fmosaic |\n| totalVI (scvi-tools) | Conditional VAE; RNA NB\u002FZINB, each protein a 2-component NB mixture (background+foreground) | Need denoised protein, principled DE, batch integration, merging different antibody panels | Tiny datasets (VAE overfits); no GPU and very large data; protein-specific background structure not captured by one per-cell factor |\n| MultiVI (scvi-tools) | Single joint VAE over RNA+ATAC(+protein); mosaic-capable, imputes missing modality | Paired+unpaired RNA\u002FATAC mixed (mosaic); want generative DE\u002FDA | \"batch\" key is the modality indicator, not sequencing batch; imputed modalities treated as measured |\n| MOFA+ (MOFA2) | Linear Bayesian group factor analysis; sparse factors, per-modality variance explained | Interpreting shared vs modality-specific axes of variation (exploratory\u002Fexplanatory) | Used for clustering\u002Fdenoising; likelihood mismatched to data; expecting batch correction within a view |\n| mojitoo | CCA across precomputed per-modality reductions; fast, parameter-free | Quick paired joint reduction from existing PCA\u002FLSI slots | No knob to down-weight a noisy modality; bounded by input reductions; paired only |\n\n## Method Decision Table (Unpaired \u002F Diagonal \u002F Mosaic)\n\n| Method | Model \u002F assumption | Use when | Fails when |\n|---|---|---|---|\n| GLUE (scglue) | Per-modality VAEs + prior feature graph (peak-near-gene); adversarial cell alignment | Unpaired diagonal scRNA + scATAC; want regulatory inference as a byproduct | Genome-build\u002Fcoordinate mismatch yields an empty guidance graph and garbage alignment; adversarial over-mixing of distinct states |\n| Seurat v5 bridge | Multiome bridge dataset = dictionary linking query modality to reference modality | Mapping a query (scATAC) onto a reference built in another modality (scRNA) | Poor\u002Fbatch-mismatched bridge propagates error; rare query-only populations mislabeled |\n| StabMap | Mosaic topology from shared features; project all cells via shortest paths | Mosaic with informative unshared features that cannot be dropped | Unshared-feature chaining compounds error per hop |\n| Cobolt \u002F scMoMaT | Generative shared latent over joint + single-modality datasets | Mosaic where a generative latent is preferred over feature chaining | DE\u002Fmarker calls made on imputed values |\n\n## ADT Normalization: CLR vs DSB\n\n| Method | What it does | Use when | Fails when |\n|---|---|---|---|\n| CLR (centered log-ratio) | Rescales compositionally; Seurat `NormalizeData(method=\"CLR\", margin=2)` | Quick, no empty droplets available; small panels | Does NOT remove background; geometric-mean denominator distorted by saturating high-abundance ADTs |\n| DSB | Ambient correction from empty droplets + per-cell technical denoising via 2-component mixture + isotype controls | Raw\u002Funfiltered matrix available (needs empty droplets); want background removed before embedding | No empty droplets retained; protein-specific non-specific binding (one per-cell factor under\u002Fover-corrects); no clearly bimodal proteins |\n\nSeurat's CLR margin is genuinely ambiguous across versions (margin=2 = per-feature is the WNN-tutorial recommendation for large panels); verify with `?NormalizeData` on the installed version.\n\n## CITE-seq: Denoise ADT, Then Joint Embed (Seurat)\n\n**Goal:** Remove ADT background with DSB before WNN, because WNN does not denoise protein.\n\n**Approach:** Estimate ambient from empty droplets and per-cell technical noise from a mixture plus isotype controls, then feed denoised ADT into the standard PCA -> WNN flow.\n\n```r\nlibrary(dsb)\nlibrary(Seurat)\n\nraw \u003C- Read10X('raw_feature_bc_matrix\u002F')           # unfiltered: contains empty droplets\ncells \u003C- Read10X('filtered_feature_bc_matrix\u002F')    # called cells\n\nadt_cells \u003C- as.matrix(cells[['Antibody Capture']])\nadt_empty \u003C- as.matrix(raw[['Antibody Capture']][, setdiff(colnames(raw[['Antibody Capture']]), colnames(adt_cells))])\n\n# isotype.control.name.vec must name the ACTUAL isotype rows (often IgG1\u002FIgG2a\u002FMouse-IgG2b-Ctrl); the regex below misses those\n# When isotypes are absent or not matched, set use.isotype.control = FALSE (keep denoise.counts = TRUE) and pass real names explicitly\nadt_dsb \u003C- DSBNormalizeProtein(\n    cell_protein_matrix = adt_cells,\n    empty_drop_matrix = adt_empty,\n    denoise.counts = TRUE,\n    use.isotype.control = TRUE,\n    isotype.control.name.vec = grep('[Ii]sotype|IgG', rownames(adt_cells), value = TRUE)\n)\n```\n\n## CITE-seq: WNN Joint Clustering (Seurat)\n\n**Goal:** Build one weighted-NN graph from denoised RNA and ADT and cluster on it.\n\n**Approach:** Reduce each modality independently (PCA on RNA, PCA on the small ADT panel), then learn per-cell modality weights and cluster\u002Fembed on the joint graph.\n\n```r\nobj[['ADT']] \u003C- CreateAssay5Object(data = adt_dsb)        # DSB output is already normalized data\nDefaultAssay(obj) \u003C- 'RNA'\nobj \u003C- NormalizeData(obj) |> FindVariableFeatures() |> ScaleData() |> RunPCA(reduction.name = 'pca')\n\nDefaultAssay(obj) \u003C- 'ADT'\nVariableFeatures(obj) \u003C- rownames(obj[['ADT']])\nobj \u003C- ScaleData(obj) |> RunPCA(reduction.name = 'apca', npcs = min(18, nrow(obj[['ADT']]) - 1))\n\n# dims.list matched to informative dims; small ADT panels saturate by ~1:18\nobj \u003C- FindMultiModalNeighbors(obj, reduction.list = list('pca', 'apca'), dims.list = list(1:30, 1:18))\nobj \u003C- FindClusters(obj, graph.name = 'wsnn', algorithm = 3)   # algorithm 3 = SLM (the tutorial choice), NOT Leiden\nobj \u003C- RunUMAP(obj, nn.name = 'weighted.nn', reduction.name = 'wnn.umap')\n\n# Inspect the per-cell weight distribution; a single dominant modality is a red flag\nVlnPlot(obj, features = 'RNA.weight', group.by = 'seurat_clusters')\n```\n\n## CITE-seq: totalVI (Python, denoise + DE in one model)\n\n**Goal:** Jointly model RNA + protein with explicit protein background, yielding a denoised latent space and foreground probabilities.\n\n**Approach:** Register a MuData object, train the conditional VAE, then read the latent representation and per-protein foreground probability.\n\n```python\nimport scvi\nimport mudata as md\n\n# mdata holds .mod['rna'] (raw counts) and .mod['prot'] (raw ADT counts)\nscvi.model.TOTALVI.setup_mudata(\n    mdata, rna_layer='counts', protein_layer=None,\n    modalities={'rna_layer': 'rna', 'protein_layer': 'prot'}\n)\nmodel = scvi.model.TOTALVI(mdata)\nmodel.train()\n\nmdata.obsm['X_totalVI'] = model.get_latent_representation()\nfg = model.get_protein_foreground_probability()        # 1 - background mixing weight per protein per cell\ndenoised_rna, denoised_prot = model.get_normalized_expression()\n```\n\n## Multiome (RNA + ATAC, same cell): Native Pipelines, Then Join\n\n**Goal:** Process each modality in its own statistics before joining, because RNA and ATAC have incompatible distributions.\n\n**Approach:** PCA on RNA, TF-IDF + LSI on ATAC (drop depth-correlated components), then WNN. See scatac-analysis for ATAC QC and the binarization\u002Fdepth-component caveats.\n\n```r\nlibrary(Signac)\nDefaultAssay(obj) \u003C- 'RNA'\nobj \u003C- NormalizeData(obj) |> FindVariableFeatures() |> ScaleData() |> RunPCA()\n\nDefaultAssay(obj) \u003C- 'ATAC'\nobj \u003C- RunTFIDF(obj) |> FindTopFeatures(min.cutoff = 'q0') |> RunSVD()\nDepthCor(obj)                                          # diagnose which LSI components track depth\n\n# dims = 2:30 drops LSI_1 ONLY if DepthCor confirms it tracks depth (usually true, not guaranteed)\nobj \u003C- FindMultiModalNeighbors(obj, reduction.list = list('pca', 'lsi'), dims.list = list(1:30, 2:30))\nobj \u003C- RunUMAP(obj, nn.name = 'weighted.nn', reduction.name = 'wnn.umap')\nobj \u003C- FindClusters(obj, graph.name = 'wsnn', algorithm = 3)\n```\n\nMerging multiome datasets requires a common peak set: re-quantify all cells against unified peaks, or peak-boundary differences manufacture spurious batch structure. The ATAC gene-activity matrix is an approximation, not measured RNA; do not conflate it with the RNA modality.\n\n## MOFA+ (interpretable shared\u002Fspecific factors)\n\n**Goal:** Decompose modalities into shared latent factors with per-modality variance explained.\n\n**Approach:** Build a MOFA object from per-modality matrices, set likelihoods to match each data type, run, then interpret factor loadings.\n\n```python\nimport muon as mu\n\n# likelihoods must match data: gaussian for scaled RNA, bernoulli for binarized ATAC, poisson for counts\nmu.tl.mofa(mdata, n_factors=15, outfile='mofa_model.hdf5')   # writes mdata.obsm['X_mofa']\n```\n\n## Unpaired \u002F Diagonal: GLUE (Python)\n\n**Goal:** Align independent scRNA and scATAC with no shared cells via a prior feature graph.\n\n**Approach:** Configure each dataset with a count-appropriate probabilistic model, build a gene-anchored guidance graph, fit GLUE, then read aligned embeddings.\n\n```python\nimport scglue\n\nscglue.models.configure_dataset(rna, 'NB', use_highly_variable=True, use_rep='X_pca')     # NB needs RAW counts\nscglue.models.configure_dataset(atac, 'ZINB', use_highly_variable=True, use_rep='X_lsi')\ngraph = scglue.genomics.rna_anchored_guidance_graph(rna, atac)     # peak-near-gene prior; coords must share genome build\nglue = scglue.models.fit_SCGLUE({'rna': rna, 'atac': atac}, graph)\nrna.obsm['X_glue'] = glue.encode_data('rna', rna)\natac.obsm['X_glue'] = glue.encode_data('atac', atac)\n```\n\nVerify cell-type structure is preserved (not just modality overlap); adversarial alignment can over-mix distinct populations.\n\n## MuData Housekeeping\n\nAfter per-modality QC, modalities hold different cell sets; `muon.pp.intersect_obs(mdata)` before any paired analysis. Editing a modality-local `mdata.mod['rna'].obs` needs `mdata.update()` to propagate to the global `mdata.obs`. R round-trips (MuDataSeurat, zellkonverter) are lossy; plan to stay in one ecosystem.\n\n## Common Errors\n\n| Symptom | Cause | Fix |\n|---|---|---|\n| WNN clustering driven entirely by ADT | A few saturating high-variance proteins dominate the neighbor graph | Report per-cell weight distribution; down-weight or denoise ADT (DSB); re-check clustering stability |\n| \"Background\" smear in every ADT cluster | Ran WNN\u002FCLR without empty-droplet denoising | Run DSB (needs raw\u002Funfiltered matrix) or totalVI before joint embedding |\n| DSB errors \u002F nonsense output | Passed a filtered cell matrix only (no empty droplets) | Supply `empty_drop_matrix` from the raw\u002Funfiltered matrix |\n| Spurious batch structure after merging multiome | Per-dataset peak sets, not a unified set | Re-quantify all cells against one common peak set |\n| GLUE produces a blob \u002F no alignment | Guidance graph near-empty from genome-build\u002Fcoordinate mismatch | Align RNA gene coords and ATAC peaks to the same build before building the graph |\n| RNA and protein disagree for a marker | Often real post-transcriptional biology (stability, trafficking, lag), not an artifact | Do not \"correct away\"; treat single-gene discordance as informative |\n| MultiVI batch effects persist | The `batch_key` was set to the modality indicator, not sequencing batch | Add a separate covariate for the real batch |\n| DE on a modality looks too clean | Computed on imputed\u002Fgene-activity values, not measurements | Flag imputed-modality DE as model-dependent; validate against a measured modality |\n\n## Related Skills\n\n- single-cell\u002Fscatac-analysis - ATAC QC, TF-IDF\u002FLSI, gene-activity caveats for the Multiome ATAC half\n- single-cell\u002Fpreprocessing - per-modality RNA QC and normalization before integration\n- single-cell\u002Fclustering - clustering and UMAP on the joint graph\n- single-cell\u002Fbatch-integration - horizontal (same-modality, cross-sample) correction\n- single-cell\u002Fmarkers-annotation - marker-based interpretation of joint clusters\n- atac-seq\u002Fmotif-deviation - chromVAR TF activity on the Multiome ATAC modality\n- pathway-analysis\u002Fgo-enrichment - functional interpretation of modality-specific factors\n\n## References\n\nArgelaguet R, Cuomo ASE, Stegle O, Marioni JC. Computational principles and challenges in single-cell data integration. Nat Biotechnol 39(10):1202-1215 (2021).\nStoeckius M, Hafemeister C, Stephenson W, et al. Simultaneous epitope and transcriptome measurement in single cells (CITE-seq). Nat Methods 14:865-868 (2017).\nMulè MP, Martins AJ, Tsang JS. Normalizing and denoising protein expression data from droplet-based single-cell profiling (DSB). Nat Commun 13:2099 (2022).\nHao Y, Hao S, Andersen-Nissen E, et al. Integrated analysis of multimodal single-cell data (WNN). Cell 184(13):3573-3587 (2021).\nGayoso A, Steier Z, Lopez R, et al. Joint probabilistic modeling of single-cell multi-omic data with totalVI. Nat Methods 18:272-282 (2021).\nAshuach T, Gabitto MI, Koodli RV, et al. MultiVI: deep generative model for the integration of multimodal data. Nat Methods 20(8):1222-1231 (2023).\nArgelaguet R, Arnol D, Bredikhin D, et al. MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data. Genome Biol 21:111 (2020).\nCao Z-J, Gao G. Multi-omics single-cell data integration and regulatory inference with graph-linked unified embedding (GLUE). Nat Biotechnol 40(10):1458-1466 (2022).\nHao Y, Stuart T, Kowalski MH, et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis (Seurat v5 bridge). Nat Biotechnol 42:293-304 (2024).\nBredikhin D, Kats I, Stegle O. MUON: multimodal omics analysis framework. Genome Biol 23:42 (2022).\nGhazanfar S, Guibentif C, Marioni JC. Stabilized mosaic single-cell data integration using unshared features (StabMap). Nat Biotechnol 42(2):284-292 (2024).\nYin Y, et al. Characterization and decontamination of background noise in droplet-based single-cell protein expression data with DecontPro. Nucleic Acids Res 52(1):e4 (2024).\n",{"name":5,"source":24,"license":15,"tool_type":25,"description":7,"imported_at":26,"source_path":27,"primary_tool":28,"source_repository":29},"https:\u002F\u002Fgithub.com\u002FGPTomics\u002FbioSkills","mixed","2026-09-20","single-cell\u002Fmultimodal-integration\u002FSKILL.md","Seurat","GPTomics\u002FbioSkills","main","2d1750717747097b602acf30bd09a7cbd4b4c6ea","awesome_skills\u002Fbiology\u002Fbio-single-cell-multimodal-integration","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous","2026-11-09 11:45:36",{"source":8,"skillName":5,"commands":37,"prompt":47,"zipAvailable":48},[38,41,44],{"tool":39,"command":40},"npx","npx skills add Mr-Nobody-Anonymous\u002FAll-skills --skill bio-single-cell-multimodal-integration",{"tool":42,"command":43},"bunx","bunx skills add Mr-Nobody-Anonymous\u002FAll-skills --skill bio-single-cell-multimodal-integration",{"tool":45,"command":46},"pnpm","pnpm dlx skills add Mr-Nobody-Anonymous\u002FAll-skills --skill bio-single-cell-multimodal-integration","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-single-cell-multimodal-integration\u002FSKILL.md",true,[50,56,62],{"path":51,"resourceType":52,"resourceTypeDesc":53,"sizeBytes":54,"sourceHtmlUrl":55},"examples\u002Fcite_seq_analysis.py","SCRIPT","脚本",2063,"https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-single-cell-multimodal-integration\u002Fexamples\u002Fcite_seq_analysis.py",{"path":57,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":60,"sourceHtmlUrl":61},"examples\u002Fcite_seq_analysis.R","OTHER","其他",1477,"https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-single-cell-multimodal-integration\u002Fexamples\u002Fcite_seq_analysis.R",{"path":63,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":66,"sourceHtmlUrl":67},"usage-guide.md","REFERENCE","参考文档",5483,"https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-single-cell-multimodal-integration\u002Fusage-guide.md",[69,77,84,91,98,105],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":76},"01M7FFFZPC2025JQQHPHDX154J","dei-assessment","Assess diversity, equity, and inclusion with metrics. TRIGGERS - Use when user needs help with dei-assessment related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fdei-assessment\u002FSKILL.md",[],[],"2026-11-16 10:05:34",{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":76},"01M7FFFWN8H4WA7A6Y398QKB6J","dei-action-plan","Create DEI action plans with goals and initiatives. TRIGGERS - Use when user needs help with dei-action-plan related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fdei-action-plan\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":76},"01M7FFFSHTM0QZRVZ2NRJTJ0PY","culture-fit-assessment","Design culture add assessment processes. TRIGGERS - Use when user needs help with culture-fit-assessment related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fculture-fit-assessment\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":76},"01M7FFFMA12RKEQA38KHJSXXPS","cowork-to-code-bridge","Use an already-installed, independently verified cowork-to-code bridge to run narrowly approved actions on the user's own macOS, Linux, or WSL2 machine through a local file queue.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fcowork-to-code-bridge\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":76},"01M7FFFH1F6M8YRGW5V487GH1K","counter-offer-response","Draft counter-offer responses for retention. TRIGGERS - Use when user needs help with counter-offer-response related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fcounter-offer-response\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"nameZh":6,"description":108,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":109,"categories":110,"keywords":111,"publishedAt":76},"01M7FFFDGEHV8Y7J9N40HZXW52","continuous-feedback-system","Design continuous feedback systems. TRIGGERS - Use when user needs help with continuous-feedback-system related tasks.","https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fhr-people\u002Fcontinuous-feedback-system\u002FSKILL.md",[],[]]