#043 — Ordinary Least Squares (OLS) Linear Regression t/F-Test
Statistical Definition
Models a continuous response as a linear combination of predictors and tests whether regression slopes significantly differ from zero.
Bioinformatics Use Case
Modeling continuous log2 normalized expression or drug response against copy number variation (CNV) dosage.
Input Variables
1 continuous response vector (y); 1+ numeric or categorical predictor variables (X).
Application
Quantify the exact rate at which gene expression scales with copy-number amplification or drug concentration using transparent, interpretable linear slope estimates.
#044 — Logistic Regression Wald & Likelihood Ratio Test
Statistical Definition
Models the log-odds of a binary outcome as a linear function of predictors, testing predictor significance via Wald z-tests or LRT.
Bioinformatics Use Case
Predicting binary disease status (Case vs. Control) or treatment response using a multi-gene biomarker panel and clinical covariates.
Input Variables
1 binary response vector (0/1); 1+ continuous or categorical predictors (X).
Application
Turn a multi-gene panel into a clinical diagnostic classifier with calibrated odds ratios showing how each biomarker shifts patient disease risk.
#045 — Poisson Regression Rate Ratio Test
Statistical Definition
Generalized linear model with log link for modeling count data where variance equals the mean, testing incidence rate ratios.
Bioinformatics Use Case
Modeling somatic mutation counts per megabase (TMB) across patient cohorts while adjusting for sequencing coverage offset.
Input Variables
1 non-negative integer count vector (y); predictors (X); optional log-exposure offset.
Application
When analyzing mutation counts or event rates across regions of varying genomic length, Poisson regression normalizes exposure and delivers exact incidence rate ratios.
#046 — Negative Binomial GLM (DESeq2 / edgeR Core Test)
Statistical Definition
Models overdispersed sequencing read counts using a gamma-Poisson mixture, testing differential abundance via Wald or likelihood ratio tests.
Bioinformatics Use Case
Bulk RNA-seq, ATAC-seq, and ChIP-seq differential gene expression / peak accessibility analysis across biological replicates.
Input Variables
Integer count matrix (genes x samples); design matrix with condition and batch covariates.
Application
This is the gold-standard statistical engine behind DESeq2 and edgeR—modeling biological overdispersion to uncover true differentially expressed genes from RNA-seq counts.
#047 — Zero-Inflated Negative Binomial (ZINB) Test
Statistical Definition
Two-component mixture model combining a point mass at zero (dropout) with a Negative Binomial count distribution for excess-zero data.
Bioinformatics Use Case
Single-cell RNA-seq differential expression and microbiome 16S/metagenomic taxa abundance modeling with structural zeros.
Input Variables
Zero-inflated integer count vector (y); predictors for count and zero-inflation components.
Application
Single-cell and microbiome datasets contain massive technical dropouts alongside true biological zeros. ZINB separates technical sparsity from real transcriptional shifts.
#048 — Beta Regression Significance Test
Statistical Definition
Models continuous dependent variables strictly bounded in the open interval (0, 1) using a beta distribution parameterized by mean and precision.
Bioinformatics Use Case
Differential DNA methylation analysis on array beta-values and modeling immune cell-type proportions from deconvolution.
Input Variables
1 continuous response bounded in (0, 1); 1+ predictors (X).
Application
Methylation beta-values and cell-type fractions live strictly between 0 and 1, violating linear regression assumptions. Beta regression models proportions natively with zero boundary artifacts.
#049 — Linear Mixed-Effects Model (LMM) Wald / LRT
Statistical Definition
Incorporates both fixed population effects and random subject/batch effects to model hierarchical, nested, or longitudinal biological data.
Bioinformatics Use Case
Single-cell multi-donor differential expression (pseudobulk/cell-level) accounting for donor-specific random intercepts.
Input Variables
Continuous response (y); fixed-effect predictors (X); random-effect grouping IDs (Z).
Application
Treating thousands of single cells from the same patient as independent replicates creates massive false positives. Linear Mixed Models properly account for donor-level clustering.
#050 — Generalized Estimating Equations (GEE) Wald Test
Statistical Definition
Estimates population-averaged marginal effects for correlated binary or count outcomes using a working correlation structure.
Bioinformatics Use Case
Longitudinal clinical trials tracking binary biomarker positivity across repeated patient visits with missing timepoints.
Input Variables
Repeated binary/count outcome (y); predictors (X); cluster/subject ID.
Application
When clinical patients miss scheduled visits or have correlated binary outcomes over time, GEE delivers robust population-level treatment effect estimates.
#051 — Quantile Regression Significance Test
Statistical Definition
Models conditional quantiles (e.g., median 50th or upper 90th percentile) rather than the conditional mean, robust to outliers and heteroscedasticity.
Bioinformatics Use Case
Modeling upper-tail viral load or mutation rate drivers where predictors affect extreme responders rather than average patients.
Input Variables
Continuous response (y); predictors (X); target quantile tau in (0, 1).
Application
Sometimes a biomarker doesn't shift the average patient—it drives extreme high-risk outliers. Quantile regression lets you model and test the exact 90th percentile of patient risk.
#052 — Nested Model Likelihood Ratio Test (LRT)
Statistical Definition
Compares the goodness-of-fit of a full statistical model against a reduced nested model using twice the difference in log-likelihoods.
Bioinformatics Use Case
Testing whether adding a 3-gene interaction signature significantly improves clinical outcome prediction beyond standard TNM staging.
Input Variables
Fitted reduced model (M0) and nested full model (M1) on identical observations.
Application
Every client wants proof that your novel biomarker adds predictive value beyond existing clinical standards. The Likelihood Ratio Test provides that exact statistical proof.
#053 — Firth Penalized Logistic Regression Test
Statistical Definition
Removes first-order bias in maximum likelihood estimation of logistic regression, resolving complete or quasi-complete separation.
Bioinformatics Use Case
Rare variant or somatic driver mutation association studies where all mutant carriers are cases and zero controls carry the variant.
Input Variables
Binary outcome (0/1); sparse or separated binary/continuous predictors (X).
Application
When a rare mutation appears exclusively in cancer patients and zero controls, standard logistic regression crashes with infinite odds ratios. Firth regression yields finite, valid p-values.
#054 — Generalized Additive Model (GAM) Smooth Term Test
Statistical Definition
Models non-linear continuous relationships using penalized regression splines and tests whether smooth trajectory terms differ from zero.
Bioinformatics Use Case
Modeling non-linear gene expression dynamics along single-cell pseudotime trajectories (tradeSeq / monocle workflows).
Input Variables
Continuous/count response (y); continuous pseudotime or spatial coordinate (t).
Application
Cells don't differentiate in straight lines. GAM spline tests pinpoint the exact genes curving up or down along single-cell developmental traje