#089 — Granger Causality F-Test [089_Granger_Causality_F_Test.ipynb]
Statistical Definition
Tests whether lagged values of one time-series improve the forecasting of another time-series beyond its own autoregressive history.
Bioinformatics Use Case
Inferring directed transcriptional regulatory cascades from dense longitudinal RNA-seq or metabolomics time-course experiments.
Input Variables
2 stationary longitudinal numeric vectors (x, y) sampled across T timepoints; lag order p.
Application
Correlation can't tell which gene activates which. Granger Causality uses longitudinal time lags to infer directed regulatory control across dynamic cellular time-courses.
#090 — Augmented Dickey-Fuller (ADF) Stationarity Test
Statistical Definition
Tests for the presence of a unit root in a time series to determine whether longitudinal biological signals are stationary or drifting.
Bioinformatics Use Case
Pre-processing continuous physiological telemetry (CGM glucose, heart-rate variability) or longitudinal microbiome trajectories prior to modeling.
Input Variables
1 longitudinal continuous vector (y) of length T.
Application
Trend artifacts ruin longitudinal biomarker models. ADF certifies whether a patient's time-series signal is stationary or requires differencing before downstream inference.
#091 — Ljung-Box Autocorrelation Portmanteau Test
Statistical Definition
Tests whether a group of autocorrelations up to lag h in a time series or model residuals are jointly zero.
Bioinformatics Use Case
Diagnosing residual temporal autocorrelation in longitudinal drug pharmacokinetics or circadian gene expression models.
Input Variables
Vector of time-series observations or model residuals; maximum lag h.
Application
Leftover temporal correlation inflates false-positive rates in longitudinal studies. Ljung-Box certifies that your model residuals are clean white noise.
#092 — Moran's I Spatial Autocorrelation Test
Statistical Definition
Quantifies spatial clustering versus dispersion of a continuous feature across 2D/3D coordinates using a spatial weights matrix.
Bioinformatics Use Case
Spatial transcriptomics (10x Visium / Slide-seq / MERFISH) detection of spatially variable genes (SVGs) across tissue architecture.
Input Variables
Continuous gene expression vector (x); 2D spatial coordinates (coords) or spatial weight matrix (W).
Application
In spatial transcriptomics, knowing "where" a gene is expressed is everything. Moran's I pinpoints spatially organized genes and immune niches across tumor tissue slides.
#093 — Ripley's K / L-Function Spatial Point Pattern Test
Statistical Definition
Evaluates whether 2D point locations exhibit spatial clustering, complete spatial randomness (CSR), or inhibition across distance radii r.
Bioinformatics Use Case
Multiplex immunofluorescence (mIF) / IMC analysis of CD8+ T-cell clustering or tumor-immune spatial exclusion.
Input Variables
2D coordinates (x, y) of cell centroids within a tissue region of area A.
Application
Are cytotoxic T-cells infiltrating tumor nests or trapped outside the tumor border? Ripley's K-function statistically proves spatial immune clustering or exclusion.
#094 — Benjamini-Hochberg False Discovery Rate (FDR) Procedure
Statistical Definition
Step-up multiple-testing correction controlling the expected proportion of false discoveries among rejected hypotheses under independence or positive dependence.
Bioinformatics Use Case
Mandatory multiple-testing adjustment across 20,000 genes in RNA-seq, 850,000 CpG sites in methylation, or thousands of metabolites.
Input Variables
Vector of M raw unadjusted p-values; target FDR threshold q (e.g., 0.05).
Application
Testing 20,000 genes without multiple-testing correction guarantees 1,000 false positives by pure chance. Benjamini-Hochberg FDR delivers publication-grade q-values that protect your R&D budget.
#095 — Benjamini-Yekutieli Dependent FDR Procedure
Statistical Definition
Conservative step-up FDR procedure guaranteed to control the false discovery rate under arbitrary negative or complex dependency structures.
Bioinformatics Use Case
Multiple-testing correction across negatively correlated compositional microbiome taxa or competitive pathway scores.
Input Variables
Vector of M raw unadjusted p-values; target FDR threshold q.
Application
Compositional microbiome and metabolomics features exhibit negative correlations that break standard FDR. Benjamini-Yekutieli guarantees strict false-discovery control under any correlation structure.
#096 — Holm-Bonferroni Step-Down Family-Wise Error Rate (FWER) Test
Statistical Definition
Uniformly more powerful step-down alternative to Bonferroni that strictly controls the probability of making even a single false discovery.
Bioinformatics Use Case
Confirmatory clinical biomarker validation panels or post-hoc pairwise contrasts where zero false positives are permitted.
Input Variables
Vector of M raw p-values; target FWER alpha.
Application
For late-stage confirmatory biomarker validation where even one false positive is unacceptable, Holm-Bonferroni controls family-wise error with strictly higher power than standard Bonferroni.
#097 — Tukey's Honestly Significant Difference (HSD) Post-Hoc Test
Statistical Definition
Pairwise multiple-comparison test using the Studentized Range distribution to identify which specific group pairs differ after a significant ANOVA.
Bioinformatics Use Case
Pinpointing which exact drug doses (10mg, 50mg, 100mg) significantly outperform placebo following a multi-arm ANOVA.
Input Variables
Continuous response vector (y); categorical grouping factor (k >= 3 levels).
Application
A significant ANOVA tells you *something* changed across your treatment arms; Tukey's HSD tells you *which exact doses* beat placebo while controlling family-wise error.
#098 — Dunn's Non-Parametric Post-Hoc Multiple Comparison Test
Statistical Definition
Non-parametric pairwise post-hoc test using average rank differences with tied-rank variance adjustment following a Kruskal-Wallis test.
Bioinformatics Use Case
Identifying specific microbiome enterotypes or skewed clinical cohorts that differ after a significant Kruskal-Wallis test.
Input Variables
Continuous/ordinal response vector (y); categorical factor (k >= 3 levels).
Application
The non-parametric companion to Kruskal-Wallis: Dunn's post-hoc test pinpoints exact pairwise cohort differences in skewed clinical or microbiome data.
#099 — Exact / Monte Carlo Permutation Test of Location
Statistical Definition
Distribution-free resampling test evaluating any custom test statistic by randomly shuffling group labels to construct an exact empirical null distribution.
Bioinformatics Use Case
Small-sample multi-omics or custom bioinformatics metrics (e.g., median fold-change or trimmed mean difference) with unknown analytical distributions.
Input Variables
2 numeric vectors (x1, x2); custom statistic function; number of permutations B.
Application
When complex bioinformatics metrics break standard textbook formulas, Monte Carlo permutation testing builds an exact, assumption-free empirical null distribution directly from your data.
#100 — BCa Bootstrap Confidence Interval & Hypothesis Test
Statistical Definition
Bias-Corrected and Accelerated (BCa) non-parametric bootstrap resampling for estimating accurate confidence intervals and p-values of complex estimators.
Bioinformatics Use Case
Quantifying uncertainty around complex bioinformatics biomarkers such as Shannon diversity, AUC, IC50, or pathway enrichment ratios.
Input Variables
Sample data vector/matrix; target estimator function theta(x); bootstrap replicates B.
Application
Complex clinical indices like IC50 or Shannon entropy lack simple variance formulas. BCa Bootstrap provides bias-corrected confidence intervals that stand up to regulatory s