[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EG5GM99TH3GTASKDFZ8VTP":3,"skill-same:01M7EG5GM99TH3GTASKDFZ8VTP":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},"01M7EG5GM99TH3GTASKDFZ8VTP","bio-temporal-genomics-temporal-grn",null,"Infers directed, time-delayed gene regulatory edges from BULK time-series expression using Granger causality (statsmodels VAR F-test), dynGENIE3 (tree ensembles regressing ODE-derived derivatives; Random Forests by default, Extra-Trees optional), and dynamic Bayesian networks (bnlearn). Use when the output is a RANKED HYPOTHESIS list for perturbation validation, not validated causal edges; deciding Granger vs dynGENIE3 vs DBN by timepoint count and linearity; sizing maxlag against the n>3*maxlag+1 degrees-of-freedom floor; handling stationarity\u002Fdifferencing before Granger; restricting regulators to known TFs; and comparing network rewiring across conditions at matched edge density. Not for single-cell pseudotime GRNs (see gene-regulatory-networks\u002Fscenic-regulons) or static co-expression (see gene-regulatory-networks\u002Fcoexpression-networks).","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-temporal-genomics-temporal-grn\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: bio-temporal-genomics-temporal-grn\ndescription: Infers directed, time-delayed gene regulatory edges from BULK time-series expression using Granger causality (statsmodels VAR F-test), dynGENIE3 (tree ensembles regressing ODE-derived derivatives; Random Forests by default, Extra-Trees optional), and dynamic Bayesian networks (bnlearn). Use when the output is a RANKED HYPOTHESIS list for perturbation validation, not validated causal edges; deciding Granger vs dynGENIE3 vs DBN by timepoint count and linearity; sizing maxlag against the n>3*maxlag+1 degrees-of-freedom floor; handling stationarity\u002Fdifferencing before Granger; restricting regulators to known TFs; and comparing network rewiring across conditions at matched edge density. Not for single-cell pseudotime GRNs (see gene-regulatory-networks\u002Fscenic-regulons) or static co-expression (see gene-regulatory-networks\u002Fcoexpression-networks).\ntool_type: mixed\nprimary_tool: statsmodels\nsource: \"https:\u002F\u002Fgithub.com\u002FGPTomics\u002FbioSkills\"\nsource_repository: \"GPTomics\u002FbioSkills\"\nsource_path: \"temporal-genomics\u002Ftemporal-grn\u002FSKILL.md\"\nlicense: \"MIT\"\nimported_at: \"2026-09-20\"\n---\n\n## Version Compatibility\n\nReference examples tested with: statsmodels 0.14+, numpy 1.26+, pandas 2.2+, dynGENIE3 (GitHub vahuynh\u002FdynGENIE3), bnlearn 4.9+, R 4.x\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\nNote: bulk time-series GRN inference is low-precision and assumption-heavy. Every edge is a HYPOTHESIS. Results are dominated by the sampling design (interval vs the minutes-scale of transcription, number of timepoints, replicate count), not by the algorithm. A tiny p-value from 6-12 timepoints is not evidence of regulation.\n\n# Temporal Gene Regulatory Network Inference\n\n**\"Infer causal regulatory relationships from my time-series expression data\"** -> Rank directed, time-delayed TF->target edges from bulk temporal expression, to prioritize perturbation experiments.\n- Python: `statsmodels.tsa.stattools.grangercausalitytests()` (VAR F-test on predictive precedence)\n- R: `dynGENIE3::dynGENIE3()` (tree ensembles on ODE-derived derivatives); `bnlearn::hc()` + `boot.strength()` (dynamic Bayesian network)\n\n## The governing principle: inference produces ranked HYPOTHESES, not validated causal edges\n\nBulk temporal GRN inference turns a time course into a ranked list of candidate directed edges whose only honest downstream use is prioritizing perturbation experiments (knockdown \u002F overexpression + re-measure). Two hard facts set the ceiling and must be stated up front, not buried.\n\n1. Granger is PREDICTIVE precedence, not mechanism. It tests whether past X improves prediction of future Y, which is neither necessary nor sufficient for regulation. It collapses in three routine biological situations:\n   - Unobserved common driver (confounding). An unmeasured TF, or a shared circadian\u002Fcell-cycle oscillation driving hundreds of genes, makes X \"Granger-cause\" Y with zero direct regulation. Pairwise methods are structurally blind to this; a shared sinusoid manufactures dense, entirely spurious directed structure whose lags are just phase offsets.\n   - Sampling coarser than the regulation timescale (aliasing). Transcription acts in minutes; bulk courses are sampled every 1-6 h. When the interval exceeds the regulatory delay, cause and effect land in the same sampled timepoint and directionality becomes unidentifiable. No statistic recovers information the sampling threw away.\n   - Non-stationarity. VAR-Granger assumes weak stationarity, but the interesting biology (a stimulus response, a developmental transient, a monotone induction) IS the non-stationary trend, and differencing it away removes the signal (see the differencing dilemma below).\n2. Community benchmarks put a LOW ceiling on precision and no single method wins. DREAM5 (Marbach 2012 *Nat Methods* 9:796) evaluated 30+ methods and found time-series network inference is low-precision, no method is best across datasets, and the robust win is the \"wisdom of crowds\": integrating independent methods beats any one. Prior information (restricting regulators to known TFs) is the other reliable lever.\n\nOperational consequence: restrict regulators to annotated TFs, run more than one method, keep edges recovered by >=2 methods and stable across replicate series, match density before comparing conditions, and hand the top edges to perturbation. This skill is bounded to BULK real-clock-time data; single-cell pseudotime GRN is a different problem (gene-regulatory-networks\u002Fscenic-regulons).\n\n## Method selection\n\n| Method | Models | Best when | Fails when |\n|--------|--------|-----------|------------|\n| Granger (statsmodels) | Bivariate VAR; F-test restricted vs unrestricted | Enough timepoints (n comfortably > 3*maxlag+1); a small a-priori TF->target set; roughly linear, stationary-after-differencing series | 6-12 timepoints (no residual DoF -> no power); genome-wide pairwise (confounding + O(TF*target) tests); saturating\u002Fswitch-like regulation (linear only) |\n| dynGENIE3 (R) | Semi-ODE: trees regress dx\u002Fdt on regulator expression | Non-linear \u002F combinatorial regulation; multiple replicates and reasonably dense sampling; a curated regulator list | Sparse or unevenly-spaced timepoints (finite-difference derivative is garbage); calibrated significance is required (it gives a RANKING, no p-values) |\n| DBN (bnlearn) | Unrolled first-order Markov Bayesian network across slices | Feedback loops matter (autoregulation, negative feedback); a pre-filtered set of tens-to-low-hundreds of nodes; edge-confidence needed | Genome-wide (super-exponential DAG search); delays longer than one sampling interval (first-order Markov); tiny samples (CI\u002Fscore tests underpowered) |\n\nMethodology evolves; verify current best practice against each tool's latest documentation before committing to one. The defensible default is to run more than one and intersect.\n\n## Granger causality (Python \u002F statsmodels)\n\n**Goal:** Rank TF->target pairs by whether past TF expression improves prediction of future target expression, with honest multiple-testing control.\n\n**Approach:** Difference all genes uniformly to approach stationarity, select a single lag per pair by BIC (so the reported p-value is not the best-of-several), run ONE F-test at that lag, then BH-correct across pairs. Test only TF->target pairs to shrink the family and encode the TF prior.\n\nThe F-test compares an unrestricted VAR (Y on its own lags AND X's lags) to a restricted model (Y on its own lags only); statsmodels reports it as `ssr_ftest`, matching R's `lmtest::grangertest`. Two constraints dominate:\n\n- Degrees-of-freedom floor. After lagging, `n_eff = n - maxlag` rows fit `2*maxlag+1` parameters, so the test is only defined for `n > 3*maxlag + 1`, and barely-defined means no power. With n=8 and maxlag=2 the F-test has ~1 residual DoF: a coin flip. This, not compute, is why genome-wide pairwise Granger fails. Prefer maxlag=1 on short courses.\n- Lag selection is itself a multiple test. Taking the minimum p-value over lags 1..maxlag and reporting it as a single test inflates significance. Fix by selecting one lag a priori, or by BIC (below), or by Bonferroni across lags before the across-pairs BH.\n\n```python\nimport numpy as np\nimport pandas as pd\nfrom statsmodels.tsa.api import VAR\nfrom statsmodels.tsa.stattools import grangercausalitytests\nfrom statsmodels.stats.multitest import multipletests\n\n# expr_df: genes x timepoints DataFrame; columns MUST be in temporal order.\n# Difference uniformly to approach stationarity. Uniform (not per-gene) differencing\n# keeps every series on the same footing: mixing I(0) and differenced I(1) series in one\n# VAR corrupts the F-test reference distribution. Cost: over-differencing already-stationary\n# genes. The deeper tradeoff: differencing removes the trend that CARRIES the regulatory\n# signal, so on short courses prefer maxlag=1 over aggressive differencing.\nexpr_diff = expr_df.diff(axis=1).iloc[:, 1:]\n\ntf_genes = ['TF1', 'TF2', 'TF3']\ntarget_genes = ['geneA', 'geneB', 'geneC']\nmaxlag = 1  # short courses have ~no DoF beyond lag 1 (need n > 3*maxlag+1)\n\ndef granger_pvalue(pair_data, maxlag):\n    # column 0 = response Y (target), column 1 = predictor X (TF): tests X -> Y.\n    # Select ONE lag by BIC, then run a SINGLE test at it -> avoids the min-p-over-lags\n    # multiple test. Guard BIC=0 (no lag structure) up to 1.\n    lag = max(1, int(VAR(pair_data).select_order(maxlag).bic))\n    res = grangercausalitytests(pair_data, maxlag=[lag])  # list -> tests only this lag\n    return res[lag][0]['ssr_ftest'][1], lag  # (p_value, lag)\n\nrecords = []\nfor tf in tf_genes:\n    for target in target_genes:\n        if tf == target:\n            continue\n        pair = np.column_stack([expr_diff.loc[target].values, expr_diff.loc[tf].values])\n        p, lag = granger_pvalue(pair, maxlag)\n        records.append({'tf': tf, 'target': target, 'p_value': p, 'lag': lag})\n\nresults_df = pd.DataFrame(records)\n# multipletests default is Holm-Sidak, NOT BH; force fdr_bh explicitly.\nresults_df['q_value'] = multipletests(results_df['p_value'], method='fdr_bh')[1]\nsignificant = results_df[results_df['q_value'] \u003C 0.05].sort_values('q_value')\n```\n\nPairwise Granger cannot separate direct regulation from a chain X->Z->Y or a fork Z->{X,Y}. The correct fix is conditional (multivariate) Granger, conditioning on all other regulators' lags, but that explodes the parameter count and is infeasible at transcriptomic sample sizes. Label pairwise output as a CONFOUNDED candidate set, not direct interactions.\n\n## dynGENIE3 (R)\n\n**Goal:** Rank regulator->target edges non-linearly by how much a regulator's current expression predicts a target's temporal derivative.\n\n**Approach:** dynGENIE3 models each gene as `dx_i\u002Fdt = f_i(x) - alpha_i * x_i`, estimates `dx_i\u002Fdt` by finite differences between consecutive timepoints, and trains a tree ensemble to regress that derivative-plus-decay target on candidate-regulator expression; summed variable importance becomes the edge weight.\n\n```r\nlibrary(dynGENIE3)\n\n# TS.data: list of genes x timepoints matrices (one per replicate\u002Fseries).\n# time.points: matching list of time vectors (real deltas -> handles uneven spacing).\nexpr_list \u003C- list(as.matrix(expr_series1), as.matrix(expr_series2), as.matrix(expr_series3))\ntime_list \u003C- list(c(0, 4, 8, 12, 24, 48), c(0, 4, 8, 12, 24, 48), c(0, 4, 8, 12, 24, 48))\n\n# Restrict regulators to known TFs (AnimalTFDB \u002F PlantTFDB). This helps TWICE: fewer\n# features searched per split (faster) AND a non-TF can never be reported as a regulator\n# (higher precision). Single highest-yield precision lever.\ntf_indices \u003C- which(rownames(expr_list[[1]]) %in% tf_names)\n\n# tree.method DEFAULTS to 'RF' (Random Forests). Extra-Trees is opt-in: tree.method='ET'\n# (the config GENIE3 used to win DREAM4). alpha='from.data' (default) estimates per-gene\n# mRNA decay from the data; pass a numeric vector to inject measured half-lives (4sU\u002FBRIC-seq).\nres \u003C- dynGENIE3(TS.data = expr_list, time.points = time_list, regulators = tf_indices)\n\n# get.link.list (DOT form) is the dynGENIE3 function. The camelCase getLinkList belongs to\n# the separate Bioconductor GENIE3 package -- do not swap them.\nlink_list \u003C- get.link.list(res$weight.matrix, report.max = 1000)\n```\n\nTwo properties gate interpretation:\n- The weight matrix is a RANKING with no null, no p-value, no calibrated threshold. A \"top edge\" is top only relative to the others in this run; thresholding by rank (top-K) is unavoidably arbitrary. This is why cross-method agreement and stability matter more here than anywhere.\n- Finite-difference derivatives amplify noise. With few, unevenly-spaced timepoints (0,4,8,12,24,48 h is typical) each `dx\u002Fdt` rests on one noisy pair and late wide intervals blur short-timescale regulation into a single slope. More REPLICATES (independent derivative samples averaging the noise down) help far more than adding one or two timepoints.\n\n## Dynamic Bayesian networks (R \u002F bnlearn)\n\n**Goal:** Learn a directed network that can represent feedback, with bootstrap edge confidence, over a pre-filtered gene set.\n\n**Approach:** Unroll time into t-1 and t slices and allow edges only from t-1 to t; because A_{t-1}->B_t and B_{t-1}->A_t both point forward, the unrolled graph is acyclic even though the biology has an A\u003C->B feedback loop. So DBNs represent feedback that static Bayesian networks (which must be DAGs) structurally cannot -- the main reason to reach for one. The cost: it is first-order Markov (state at t depends only on t-1; longer delays need t-2\u002Ft-3 slices) and the super-exponential DAG search caps realistic inference at tens-to-low-hundreds of nodes, never genome-wide.\n\n```r\nlibrary(bnlearn)\n\n# Build the 2-slice frame: columns _t1 (predictors at t-1) and _t (response at t).\nn_t \u003C- ncol(expr_mat)\nlagged_df \u003C- data.frame(\n    t(expr_mat[, 2:n_t]),      # response slice t\n    t(expr_mat[, 1:(n_t - 1)]) # predictor slice t-1\n)\ncolnames(lagged_df) \u003C- c(paste0(rownames(expr_mat), '_t'),\n                         paste0(rownames(expr_mat), '_t1'))\n\n# Constrain edges to t-1 -> t so the learned graph is a proper DBN transition model.\nnodes_t  \u003C- paste0(rownames(expr_mat), '_t')\nnodes_t1 \u003C- paste0(rownames(expr_mat), '_t1')\nblacklist \u003C- rbind(\n    expand.grid(from = nodes_t, to = nodes_t1),   # forbid t -> t-1 (backward in time)\n    expand.grid(from = nodes_t1, to = nodes_t1)   # forbid within-slice t-1 edges\n)\n\n# score='bic-g': Gaussian BIC; penalizes parameters, guarding the tiny sample against\n# overfit. Gaussian assumes linear-Gaussian dependencies (misses threshold logic, like\n# Granger); discretizing captures nonlinearity but needs data you do not have on short\n# courses. hc is greedy -> trust boot.strength, not one DAG.\nboot_res \u003C- boot.strength(lagged_df, R = 200, algorithm = 'hc',\n                          algorithm.args = list(score = 'bic-g', blacklist = blacklist))\n\n# strength = fraction of bootstraps containing the arc; direction = fraction of those\n# oriented the stated way. direction >= 0.5 is a COIN FLIP -- require >= 0.8 for a\n# confidently oriented edge. bnlearn can also compute a data-driven strength threshold:\nthr \u003C- attr(boot_res, 'threshold')  # data-driven threshold lives on the bn.strength object, a principled alternative to hand-picked 0.7\nconfident \u003C- boot_res[boot_res$strength >= max(0.7, thr) & boot_res$direction >= 0.8, ]\n```\n\n## Comparing networks across conditions\n\n**Goal:** Identify genuine rewiring between two conditions, not artifacts of threshold choice.\n\n**Approach:** Edge-set differences are dominated by density mismatch and near-threshold flips unless controlled. Compare at MATCHED edge density (top-K from each, same K), and only call an edge gained\u002Flost if it is present-and-bootstrap-stable in one condition and absent-and-stable in the other.\n\n```python\ndef top_k_edges(edge_df, k):\n    return set(map(tuple, edge_df.sort_values('weight', ascending=False)\n                   .head(k)[['tf', 'target']].values))\n\nk = min(len(edges_a), len(edges_b))  # density-match BEFORE comparing\nset_a, set_b = top_k_edges(edges_a, k), top_k_edges(edges_b, k)\njaccard = len(set_a & set_b) \u002F len(set_a | set_b) if (set_a | set_b) else 0.0\ngained, lost = set_b - set_a, set_a - set_b  # keep only bootstrap-stable ones\n```\n\nJaccard heuristics (\u003C 0.3 rewired, > 0.7 conserved) are uncalibrated and, without density-matching, mostly measure the threshold rather than biology -- present them as rough anchors only after matching.\n\n## What experts do instead of trusting one method\n\n- Prior-constrain regulators to annotated TFs (dynGENIE3 `regulators=`; Granger test only TF->target; DBN whitelist\u002Fblacklist). Highest-yield, cheapest precision lever.\n- Ensemble across methods; edges recovered by >=2 orthogonal methods are the ones worth an experiment (Marbach 2012's wisdom-of-crowds result).\n- Require replication across independent time series; bootstrap-subsample and re-rank to separate reproducible edges from artifacts.\n- Treat the output as a prioritized hypothesis list for perturbation. Nothing in bulk inference validates an edge; only perturbation does.\n\n## Common Errors\n\n| Symptom | Cause | Fix |\n|---------|-------|-----|\n| `grangercausalitytests(..., verbose=False)` raises FutureWarning | `verbose` deprecated since statsmodels 0.14, slated for removal | Drop the argument; index the returned dict (`res[lag][0]['ssr_ftest'][1]`) |\n| Granger q-values suspiciously optimistic | `min_p` across lags then BH is an uncorrected within-pair multiple test | Fix one lag a priori, or BIC-select one lag then run a single test, or Bonferroni across lags before BH |\n| Granger has no power \u002F errors on few timepoints | `n > 3*maxlag+1` barely met -> ~1 residual DoF | Use maxlag=1 on short courses; get more timepoints\u002Freplicates before trusting any q-value |\n| \"dynGENIE3 uses Extra-Trees\" | dynGENIE3 defaults to `tree.method='RF'` (Random Forests); ET is opt-in | Pass `tree.method='ET'` if ET is wanted, else describe it as RF |\n| dynGENIE3 edges read as calibrated | Importances have no null \u002F no p-value | Threshold by rank explicitly; validate top edges by cross-method agreement + perturbation |\n| dynGENIE3 gives garbage on sparse\u002Funeven series | Finite-difference `dx\u002Fdt` amplifies noise | Add replicates (independent derivative samples), not just one more timepoint |\n| DBN `direction >= 0.5` admits reversed edges | 0.5 = \"more often than not\" = coin-flip orientation | Require `direction >= 0.8`; consider bnlearn's data-driven strength threshold over a hand-picked 0.7 |\n| Pairwise Granger reported as direct regulation | Blind to common drivers \u002F chains; circadian oscillation fabricates dense edges | Label as confounded candidates; restrict to TF->target; intersect methods |\n| Jaccard swings wildly between conditions | Density mismatch + near-threshold flips, not biology | Match edge density (top-K each); require bootstrap-stable presence\u002Fabsence |\n| Lag structure vanishes silently | Expression columns not in temporal order | Assert timepoint ordering before any lagging |\n\n## Related Skills\n\n- gene-regulatory-networks\u002Fcoexpression-networks - Static (non-temporal) co-expression networks\n- gene-regulatory-networks\u002Fscenic-regulons - Single-cell pseudotime regulon inference (different data and assumptions)\n- gene-regulatory-networks\u002Fdifferential-networks - Condition-specific network comparison\n- differential-expression\u002Ftimeseries-de - Filter to temporally-variable genes before edge inference\n- data-visualization\u002Fnetwork-visualization - Plotting inferred networks\n\n## References\n\n- Granger CWJ. 1969. Investigating causal relations by econometric models and cross-spectral methods. *Econometrica* 37(3):424-438. Predictive-precedence definition of causality.\n- Huynh-Thu VA, Geurts P. 2018. dynGENIE3: dynamical GENIE3 for the inference of gene networks from time series expression data. *Sci Rep* 8:3384. Semi-ODE + tree-regression-on-derivative method.\n- Huynh-Thu VA, Irrthum A, Wehenkel L, Geurts P. 2010. Inferring regulatory networks from expression data using tree-based methods. *PLoS ONE* 5(9):e12776. GENIE3 tree-based variable selection.\n- Marbach D, Costello JC, Kuffner R, et al. 2012. Wisdom of crowds for robust gene network inference. *Nat Methods* 9(8):796-804. Low precision, no single method wins, community-ensemble superiority.\n- Scutari M. 2010. Learning Bayesian networks with the bnlearn R package. *J Stat Softw* 35(3):1-22. bnlearn `hc` \u002F `boot.strength` API.\n- Friedman N, Murphy K, Russell S. 1998. Learning the structure of dynamic probabilistic networks. *Proc. 14th Conf. on Uncertainty in Artificial Intelligence (UAI)*, pp. 139-147. Score-based DBN structure learning.\n","## Version Compatibility\n\nReference examples tested with: statsmodels 0.14+, numpy 1.26+, pandas 2.2+, dynGENIE3 (GitHub vahuynh\u002FdynGENIE3), bnlearn 4.9+, R 4.x\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\nNote: bulk time-series GRN inference is low-precision and assumption-heavy. Every edge is a HYPOTHESIS. Results are dominated by the sampling design (interval vs the minutes-scale of transcription, number of timepoints, replicate count), not by the algorithm. A tiny p-value from 6-12 timepoints is not evidence of regulation.\n\n# Temporal Gene Regulatory Network Inference\n\n**\"Infer causal regulatory relationships from my time-series expression data\"** -> Rank directed, time-delayed TF->target edges from bulk temporal expression, to prioritize perturbation experiments.\n- Python: `statsmodels.tsa.stattools.grangercausalitytests()` (VAR F-test on predictive precedence)\n- R: `dynGENIE3::dynGENIE3()` (tree ensembles on ODE-derived derivatives); `bnlearn::hc()` + `boot.strength()` (dynamic Bayesian network)\n\n## The governing principle: inference produces ranked HYPOTHESES, not validated causal edges\n\nBulk temporal GRN inference turns a time course into a ranked list of candidate directed edges whose only honest downstream use is prioritizing perturbation experiments (knockdown \u002F overexpression + re-measure). Two hard facts set the ceiling and must be stated up front, not buried.\n\n1. Granger is PREDICTIVE precedence, not mechanism. It tests whether past X improves prediction of future Y, which is neither necessary nor sufficient for regulation. It collapses in three routine biological situations:\n   - Unobserved common driver (confounding). An unmeasured TF, or a shared circadian\u002Fcell-cycle oscillation driving hundreds of genes, makes X \"Granger-cause\" Y with zero direct regulation. Pairwise methods are structurally blind to this; a shared sinusoid manufactures dense, entirely spurious directed structure whose lags are just phase offsets.\n   - Sampling coarser than the regulation timescale (aliasing). Transcription acts in minutes; bulk courses are sampled every 1-6 h. When the interval exceeds the regulatory delay, cause and effect land in the same sampled timepoint and directionality becomes unidentifiable. No statistic recovers information the sampling threw away.\n   - Non-stationarity. VAR-Granger assumes weak stationarity, but the interesting biology (a stimulus response, a developmental transient, a monotone induction) IS the non-stationary trend, and differencing it away removes the signal (see the differencing dilemma below).\n2. Community benchmarks put a LOW ceiling on precision and no single method wins. DREAM5 (Marbach 2012 *Nat Methods* 9:796) evaluated 30+ methods and found time-series network inference is low-precision, no method is best across datasets, and the robust win is the \"wisdom of crowds\": integrating independent methods beats any one. Prior information (restricting regulators to known TFs) is the other reliable lever.\n\nOperational consequence: restrict regulators to annotated TFs, run more than one method, keep edges recovered by >=2 methods and stable across replicate series, match density before comparing conditions, and hand the top edges to perturbation. This skill is bounded to BULK real-clock-time data; single-cell pseudotime GRN is a different problem (gene-regulatory-networks\u002Fscenic-regulons).\n\n## Method selection\n\n| Method | Models | Best when | Fails when |\n|--------|--------|-----------|------------|\n| Granger (statsmodels) | Bivariate VAR; F-test restricted vs unrestricted | Enough timepoints (n comfortably > 3*maxlag+1); a small a-priori TF->target set; roughly linear, stationary-after-differencing series | 6-12 timepoints (no residual DoF -> no power); genome-wide pairwise (confounding + O(TF*target) tests); saturating\u002Fswitch-like regulation (linear only) |\n| dynGENIE3 (R) | Semi-ODE: trees regress dx\u002Fdt on regulator expression | Non-linear \u002F combinatorial regulation; multiple replicates and reasonably dense sampling; a curated regulator list | Sparse or unevenly-spaced timepoints (finite-difference derivative is garbage); calibrated significance is required (it gives a RANKING, no p-values) |\n| DBN (bnlearn) | Unrolled first-order Markov Bayesian network across slices | Feedback loops matter (autoregulation, negative feedback); a pre-filtered set of tens-to-low-hundreds of nodes; edge-confidence needed | Genome-wide (super-exponential DAG search); delays longer than one sampling interval (first-order Markov); tiny samples (CI\u002Fscore tests underpowered) |\n\nMethodology evolves; verify current best practice against each tool's latest documentation before committing to one. The defensible default is to run more than one and intersect.\n\n## Granger causality (Python \u002F statsmodels)\n\n**Goal:** Rank TF->target pairs by whether past TF expression improves prediction of future target expression, with honest multiple-testing control.\n\n**Approach:** Difference all genes uniformly to approach stationarity, select a single lag per pair by BIC (so the reported p-value is not the best-of-several), run ONE F-test at that lag, then BH-correct across pairs. Test only TF->target pairs to shrink the family and encode the TF prior.\n\nThe F-test compares an unrestricted VAR (Y on its own lags AND X's lags) to a restricted model (Y on its own lags only); statsmodels reports it as `ssr_ftest`, matching R's `lmtest::grangertest`. Two constraints dominate:\n\n- Degrees-of-freedom floor. After lagging, `n_eff = n - maxlag` rows fit `2*maxlag+1` parameters, so the test is only defined for `n > 3*maxlag + 1`, and barely-defined means no power. With n=8 and maxlag=2 the F-test has ~1 residual DoF: a coin flip. This, not compute, is why genome-wide pairwise Granger fails. Prefer maxlag=1 on short courses.\n- Lag selection is itself a multiple test. Taking the minimum p-value over lags 1..maxlag and reporting it as a single test inflates significance. Fix by selecting one lag a priori, or by BIC (below), or by Bonferroni across lags before the across-pairs BH.\n\n```python\nimport numpy as np\nimport pandas as pd\nfrom statsmodels.tsa.api import VAR\nfrom statsmodels.tsa.stattools import grangercausalitytests\nfrom statsmodels.stats.multitest import multipletests\n\n# expr_df: genes x timepoints DataFrame; columns MUST be in temporal order.\n# Difference uniformly to approach stationarity. Uniform (not per-gene) differencing\n# keeps every series on the same footing: mixing I(0) and differenced I(1) series in one\n# VAR corrupts the F-test reference distribution. Cost: over-differencing already-stationary\n# genes. The deeper tradeoff: differencing removes the trend that CARRIES the regulatory\n# signal, so on short courses prefer maxlag=1 over aggressive differencing.\nexpr_diff = expr_df.diff(axis=1).iloc[:, 1:]\n\ntf_genes = ['TF1', 'TF2', 'TF3']\ntarget_genes = ['geneA', 'geneB', 'geneC']\nmaxlag = 1  # short courses have ~no DoF beyond lag 1 (need n > 3*maxlag+1)\n\ndef granger_pvalue(pair_data, maxlag):\n    # column 0 = response Y (target), column 1 = predictor X (TF): tests X -> Y.\n    # Select ONE lag by BIC, then run a SINGLE test at it -> avoids the min-p-over-lags\n    # multiple test. Guard BIC=0 (no lag structure) up to 1.\n    lag = max(1, int(VAR(pair_data).select_order(maxlag).bic))\n    res = grangercausalitytests(pair_data, maxlag=[lag])  # list -> tests only this lag\n    return res[lag][0]['ssr_ftest'][1], lag  # (p_value, lag)\n\nrecords = []\nfor tf in tf_genes:\n    for target in target_genes:\n        if tf == target:\n            continue\n        pair = np.column_stack([expr_diff.loc[target].values, expr_diff.loc[tf].values])\n        p, lag = granger_pvalue(pair, maxlag)\n        records.append({'tf': tf, 'target': target, 'p_value': p, 'lag': lag})\n\nresults_df = pd.DataFrame(records)\n# multipletests default is Holm-Sidak, NOT BH; force fdr_bh explicitly.\nresults_df['q_value'] = multipletests(results_df['p_value'], method='fdr_bh')[1]\nsignificant = results_df[results_df['q_value'] \u003C 0.05].sort_values('q_value')\n```\n\nPairwise Granger cannot separate direct regulation from a chain X->Z->Y or a fork Z->{X,Y}. The correct fix is conditional (multivariate) Granger, conditioning on all other regulators' lags, but that explodes the parameter count and is infeasible at transcriptomic sample sizes. Label pairwise output as a CONFOUNDED candidate set, not direct interactions.\n\n## dynGENIE3 (R)\n\n**Goal:** Rank regulator->target edges non-linearly by how much a regulator's current expression predicts a target's temporal derivative.\n\n**Approach:** dynGENIE3 models each gene as `dx_i\u002Fdt = f_i(x) - alpha_i * x_i`, estimates `dx_i\u002Fdt` by finite differences between consecutive timepoints, and trains a tree ensemble to regress that derivative-plus-decay target on candidate-regulator expression; summed variable importance becomes the edge weight.\n\n```r\nlibrary(dynGENIE3)\n\n# TS.data: list of genes x timepoints matrices (one per replicate\u002Fseries).\n# time.points: matching list of time vectors (real deltas -> handles uneven spacing).\nexpr_list \u003C- list(as.matrix(expr_series1), as.matrix(expr_series2), as.matrix(expr_series3))\ntime_list \u003C- list(c(0, 4, 8, 12, 24, 48), c(0, 4, 8, 12, 24, 48), c(0, 4, 8, 12, 24, 48))\n\n# Restrict regulators to known TFs (AnimalTFDB \u002F PlantTFDB). This helps TWICE: fewer\n# features searched per split (faster) AND a non-TF can never be reported as a regulator\n# (higher precision). Single highest-yield precision lever.\ntf_indices \u003C- which(rownames(expr_list[[1]]) %in% tf_names)\n\n# tree.method DEFAULTS to 'RF' (Random Forests). Extra-Trees is opt-in: tree.method='ET'\n# (the config GENIE3 used to win DREAM4). alpha='from.data' (default) estimates per-gene\n# mRNA decay from the data; pass a numeric vector to inject measured half-lives (4sU\u002FBRIC-seq).\nres \u003C- dynGENIE3(TS.data = expr_list, time.points = time_list, regulators = tf_indices)\n\n# get.link.list (DOT form) is the dynGENIE3 function. The camelCase getLinkList belongs to\n# the separate Bioconductor GENIE3 package -- do not swap them.\nlink_list \u003C- get.link.list(res$weight.matrix, report.max = 1000)\n```\n\nTwo properties gate interpretation:\n- The weight matrix is a RANKING with no null, no p-value, no calibrated threshold. A \"top edge\" is top only relative to the others in this run; thresholding by rank (top-K) is unavoidably arbitrary. This is why cross-method agreement and stability matter more here than anywhere.\n- Finite-difference derivatives amplify noise. With few, unevenly-spaced timepoints (0,4,8,12,24,48 h is typical) each `dx\u002Fdt` rests on one noisy pair and late wide intervals blur short-timescale regulation into a single slope. More REPLICATES (independent derivative samples averaging the noise down) help far more than adding one or two timepoints.\n\n## Dynamic Bayesian networks (R \u002F bnlearn)\n\n**Goal:** Learn a directed network that can represent feedback, with bootstrap edge confidence, over a pre-filtered gene set.\n\n**Approach:** Unroll time into t-1 and t slices and allow edges only from t-1 to t; because A_{t-1}->B_t and B_{t-1}->A_t both point forward, the unrolled graph is acyclic even though the biology has an A\u003C->B feedback loop. So DBNs represent feedback that static Bayesian networks (which must be DAGs) structurally cannot -- the main reason to reach for one. The cost: it is first-order Markov (state at t depends only on t-1; longer delays need t-2\u002Ft-3 slices) and the super-exponential DAG search caps realistic inference at tens-to-low-hundreds of nodes, never genome-wide.\n\n```r\nlibrary(bnlearn)\n\n# Build the 2-slice frame: columns _t1 (predictors at t-1) and _t (response at t).\nn_t \u003C- ncol(expr_mat)\nlagged_df \u003C- data.frame(\n    t(expr_mat[, 2:n_t]),      # response slice t\n    t(expr_mat[, 1:(n_t - 1)]) # predictor slice t-1\n)\ncolnames(lagged_df) \u003C- c(paste0(rownames(expr_mat), '_t'),\n                         paste0(rownames(expr_mat), '_t1'))\n\n# Constrain edges to t-1 -> t so the learned graph is a proper DBN transition model.\nnodes_t  \u003C- paste0(rownames(expr_mat), '_t')\nnodes_t1 \u003C- paste0(rownames(expr_mat), '_t1')\nblacklist \u003C- rbind(\n    expand.grid(from = nodes_t, to = nodes_t1),   # forbid t -> t-1 (backward in time)\n    expand.grid(from = nodes_t1, to = nodes_t1)   # forbid within-slice t-1 edges\n)\n\n# score='bic-g': Gaussian BIC; penalizes parameters, guarding the tiny sample against\n# overfit. Gaussian assumes linear-Gaussian dependencies (misses threshold logic, like\n# Granger); discretizing captures nonlinearity but needs data you do not have on short\n# courses. hc is greedy -> trust boot.strength, not one DAG.\nboot_res \u003C- boot.strength(lagged_df, R = 200, algorithm = 'hc',\n                          algorithm.args = list(score = 'bic-g', blacklist = blacklist))\n\n# strength = fraction of bootstraps containing the arc; direction = fraction of those\n# oriented the stated way. direction >= 0.5 is a COIN FLIP -- require >= 0.8 for a\n# confidently oriented edge. bnlearn can also compute a data-driven strength threshold:\nthr \u003C- attr(boot_res, 'threshold')  # data-driven threshold lives on the bn.strength object, a principled alternative to hand-picked 0.7\nconfident \u003C- boot_res[boot_res$strength >= max(0.7, thr) & boot_res$direction >= 0.8, ]\n```\n\n## Comparing networks across conditions\n\n**Goal:** Identify genuine rewiring between two conditions, not artifacts of threshold choice.\n\n**Approach:** Edge-set differences are dominated by density mismatch and near-threshold flips unless controlled. Compare at MATCHED edge density (top-K from each, same K), and only call an edge gained\u002Flost if it is present-and-bootstrap-stable in one condition and absent-and-stable in the other.\n\n```python\ndef top_k_edges(edge_df, k):\n    return set(map(tuple, edge_df.sort_values('weight', ascending=False)\n                   .head(k)[['tf', 'target']].values))\n\nk = min(len(edges_a), len(edges_b))  # density-match BEFORE comparing\nset_a, set_b = top_k_edges(edges_a, k), top_k_edges(edges_b, k)\njaccard = len(set_a & set_b) \u002F len(set_a | set_b) if (set_a | set_b) else 0.0\ngained, lost = set_b - set_a, set_a - set_b  # keep only bootstrap-stable ones\n```\n\nJaccard heuristics (\u003C 0.3 rewired, > 0.7 conserved) are uncalibrated and, without density-matching, mostly measure the threshold rather than biology -- present them as rough anchors only after matching.\n\n## What experts do instead of trusting one method\n\n- Prior-constrain regulators to annotated TFs (dynGENIE3 `regulators=`; Granger test only TF->target; DBN whitelist\u002Fblacklist). Highest-yield, cheapest precision lever.\n- Ensemble across methods; edges recovered by >=2 orthogonal methods are the ones worth an experiment (Marbach 2012's wisdom-of-crowds result).\n- Require replication across independent time series; bootstrap-subsample and re-rank to separate reproducible edges from artifacts.\n- Treat the output as a prioritized hypothesis list for perturbation. Nothing in bulk inference validates an edge; only perturbation does.\n\n## Common Errors\n\n| Symptom | Cause | Fix |\n|---------|-------|-----|\n| `grangercausalitytests(..., verbose=False)` raises FutureWarning | `verbose` deprecated since statsmodels 0.14, slated for removal | Drop the argument; index the returned dict (`res[lag][0]['ssr_ftest'][1]`) |\n| Granger q-values suspiciously optimistic | `min_p` across lags then BH is an uncorrected within-pair multiple test | Fix one lag a priori, or BIC-select one lag then run a single test, or Bonferroni across lags before BH |\n| Granger has no power \u002F errors on few timepoints | `n > 3*maxlag+1` barely met -> ~1 residual DoF | Use maxlag=1 on short courses; get more timepoints\u002Freplicates before trusting any q-value |\n| \"dynGENIE3 uses Extra-Trees\" | dynGENIE3 defaults to `tree.method='RF'` (Random Forests); ET is opt-in | Pass `tree.method='ET'` if ET is wanted, else describe it as RF |\n| dynGENIE3 edges read as calibrated | Importances have no null \u002F no p-value | Threshold by rank explicitly; validate top edges by cross-method agreement + perturbation |\n| dynGENIE3 gives garbage on sparse\u002Funeven series | Finite-difference `dx\u002Fdt` amplifies noise | Add replicates (independent derivative samples), not just one more timepoint |\n| DBN `direction >= 0.5` admits reversed edges | 0.5 = \"more often than not\" = coin-flip orientation | Require `direction >= 0.8`; consider bnlearn's data-driven strength threshold over a hand-picked 0.7 |\n| Pairwise Granger reported as direct regulation | Blind to common drivers \u002F chains; circadian oscillation fabricates dense edges | Label as confounded candidates; restrict to TF->target; intersect methods |\n| Jaccard swings wildly between conditions | Density mismatch + near-threshold flips, not biology | Match edge density (top-K each); require bootstrap-stable presence\u002Fabsence |\n| Lag structure vanishes silently | Expression columns not in temporal order | Assert timepoint ordering before any lagging |\n\n## Related Skills\n\n- gene-regulatory-networks\u002Fcoexpression-networks - Static (non-temporal) co-expression networks\n- gene-regulatory-networks\u002Fscenic-regulons - Single-cell pseudotime regulon inference (different data and assumptions)\n- gene-regulatory-networks\u002Fdifferential-networks - Condition-specific network comparison\n- differential-expression\u002Ftimeseries-de - Filter to temporally-variable genes before edge inference\n- data-visualization\u002Fnetwork-visualization - Plotting inferred networks\n\n## References\n\n- Granger CWJ. 1969. Investigating causal relations by econometric models and cross-spectral methods. *Econometrica* 37(3):424-438. Predictive-precedence definition of causality.\n- Huynh-Thu VA, Geurts P. 2018. dynGENIE3: dynamical GENIE3 for the inference of gene networks from time series expression data. *Sci Rep* 8:3384. Semi-ODE + tree-regression-on-derivative method.\n- Huynh-Thu VA, Irrthum A, Wehenkel L, Geurts P. 2010. Inferring regulatory networks from expression data using tree-based methods. *PLoS ONE* 5(9):e12776. GENIE3 tree-based variable selection.\n- Marbach D, Costello JC, Kuffner R, et al. 2012. Wisdom of crowds for robust gene network inference. *Nat Methods* 9(8):796-804. Low precision, no single method wins, community-ensemble superiority.\n- Scutari M. 2010. Learning Bayesian networks with the bnlearn R package. *J Stat Softw* 35(3):1-22. bnlearn `hc` \u002F `boot.strength` API.\n- Friedman N, Murphy K, Russell S. 1998. Learning the structure of dynamic probabilistic networks. *Proc. 14th Conf. on Uncertainty in Artificial Intelligence (UAI)*, pp. 139-147. Score-based DBN structure learning.\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","temporal-genomics\u002Ftemporal-grn\u002FSKILL.md","statsmodels","GPTomics\u002FbioSkills","main","2d1750717747097b602acf30bd09a7cbd4b4c6ea","awesome_skills\u002Fbiology\u002Fbio-temporal-genomics-temporal-grn","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-temporal-genomics-temporal-grn",{"tool":42,"command":43},"bunx","bunx skills add Mr-Nobody-Anonymous\u002FAll-skills --skill bio-temporal-genomics-temporal-grn",{"tool":45,"command":46},"pnpm","pnpm dlx skills add Mr-Nobody-Anonymous\u002FAll-skills --skill bio-temporal-genomics-temporal-grn","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-temporal-genomics-temporal-grn\u002FSKILL.md",true,[50,56,62],{"path":51,"resourceType":52,"resourceTypeDesc":53,"sizeBytes":54,"sourceHtmlUrl":55},"examples\u002Fdyngenie3_temporal_grn.R","OTHER","其他",5180,"https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-temporal-genomics-temporal-grn\u002Fexamples\u002Fdyngenie3_temporal_grn.R",{"path":57,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":60,"sourceHtmlUrl":61},"examples\u002Fgranger_causality_grn.py","SCRIPT","脚本",7376,"https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-temporal-genomics-temporal-grn\u002Fexamples\u002Fgranger_causality_grn.py",{"path":63,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":66,"sourceHtmlUrl":67},"usage-guide.md","REFERENCE","参考文档",6169,"https:\u002F\u002Fgithub.com\u002FMr-Nobody-Anonymous\u002FAll-skills\u002Fblob\u002Fmain\u002Fawesome_skills\u002Fbiology\u002Fbio-temporal-genomics-temporal-grn\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",[],[]]