Core Analysis Tools

Analysis Type

Defines the high-level methodology (e.g., Random Forest, Correlation, or GWAS Screening). Controls which downstream parameters are visible.

Number of Estimators

The number of decision trees in the ensemble model. Higher values lead to more stable models but require more memory.

Number of CV Splits

The $k$ partitions used for $k$-fold cross-validation. Typical academic baseline is 5 or 10.

Test Size

Proportion of the dataset used for validation (e.g., 0.2 represents 20% for testing).

Random State (Seed)

Integer seed value ensuring that stochastic operations (like dataset splitting) are identically reproducible across runs.

Categorical Feature Encoding

How non-numeric feature columns are converted before modeling: Exclude drops them, Integer maps each level to a number (1, 2, 3…), and One-hot expands each level into its own 0/1 column (e.g., RANCH_KINSELLA).

Correlation Method

Determines how to measure feature association: Pearson (linear), Spearman (rank), or Kendall (robust).

FDR Threshold (q-value)

The false discovery rate cutoff (Benjamini-Hochberg) for reporting significant biological findings.

Permutation Repeats

The number of shuffles performed to generate internal null distributions for feature importance significance testing.

SHAP Analysis Coverage

Enables calculation of Shapley values for model local interpretability, identifying variable contribution per sample.

Gene List Source

Input selection for enrichment. Utilizes high-ranking features from Random Forest or manually curated user lists.

Feature Subset Size (Top N)

The count of highest-ranking genes to include in ORA calculations.

Minimum Column Fraction

Fraction of numeric columns a row must be outlying in before IQR/Z-Score detection flags it. Higher values are more conservative on wide datasets.

Manual Gene List

Gene or feature identifiers typed or pasted directly — one symbol per line or comma-separated. Used when Gene List Source is "Manual Gene Input".

Biological Organism

Species context for gene ID mapping (e.g., Human vs specialized genomic backends).

GO Namespace

Specifies which Gene Ontology branch to query: Biological Process, Molecular Function, or Cellular Component.

Minimum Overlap Threshold

The minimum number of provided genes required within a pathway to trigger statistical testing.

Dataset Selection

The primary data source for the analysis. All preprocessing, statistics, and modeling are executed against the records in this file. Datasets are uploaded via the Datasets page and can be filtered by project.

Target Column (Dependent Variable)

The outcome variable (Y) to predict or correlate against. For regression models this must be a continuous numeric column; leave blank to auto-detect the first numeric column.

Normalize Data

Scales feature values before analysis so that variables measured on large ranges do not dominate distance- or variance-based models. Recommended when columns mix different units or magnitudes.

Pearson Correlation

Univariate screen measuring the linear association between each feature and the target. Fast and interpretable; assumes roughly normal distributions.

Spearman Correlation

Rank-based univariate screen for monotonic (not necessarily linear) relationships. More robust to outliers and skewed data than Pearson.

ANOVA F-test

Compares feature means across groups defined by a categorical target to flag variables that differ significantly between classes or conditions.

Missing Values

How empty cells are handled before modeling: keep them, or impute each column with its mean (default), median (robust to outliers), or mode (for categorical data).

Normalization

Rescales feature magnitudes: min-max maps each column to the 0–1 range; z-score standardises to zero mean and unit variance.

Outlier Detection

The algorithm used to flag anomalous samples: IQR (simple rule), IsolationForest, LOF, EllipticEnvelope, OneClassSVM, or DBSCAN.

Outlier Handling

What to do with detected outliers: remove the rows, keep them, winsorize extreme values to percentile bounds, or impute with the column median/mean.

Outlier Timing

When outlier handling is applied relative to the train/test split. Training only avoids leaking information from test samples into the decision.

Log Transform

Applies log(x) to numeric features, compressing right-skewed distributions common in count and concentration data.

Sqrt Transform

Applies the square-root transform — a milder variance-stabilising correction than log, suited to moderately skewed count data.

Reciprocal Transform

Applies 1/x, strongly shrinking large values and spreading small ones. Useful for extremely right-skewed positive data.

Power Transform (Yeo-Johnson)

Estimates an optimal power transform per feature to approximate a normal distribution; unlike Box-Cox it supports zero and negative values.

Quantile Transform

Maps each feature to a uniform or normal distribution via its empirical quantiles. Robust to any input distribution but destroys relative distances between values.

Polynomial Features

Generates interaction and power terms between features up to the chosen degree, letting the model capture non-linear relationships (e.g., x₁·x₂, x₁²).

Polynomial Degree

The maximum power of generated polynomial terms. Higher degrees grow the feature count rapidly and risk overfitting.

Spline Transform

Expands features into piecewise polynomial basis functions so linear models can fit smooth non-linear trends such as dose-response curves.

KBins Discretize

Buckets continuous features into discrete intervals, converting numeric variables into categorical bins.

Pairwise Ratios

Creates ratio features between column pairs, exposing proportional or stoichiometric relationships that absolute values hide.

Feature Selection Method

Strategy for reducing the feature set before modeling: variance threshold, correlation filter, recursive feature elimination (RFE), K-Best scoring, percentile, FDR, or model-based selection.

K Features

The number of top-ranked features retained by the selection method.

Scoring Function

The metric used to rank features during selection: R², F-regression, or mutual information.

Task Type

Regression predicts a continuous numeric outcome; classification predicts discrete categories. The choice determines which model list and metrics apply.

Model

The machine learning algorithm to train. Random Forest is a robust default for tabular omics data; linear models suit well-understood linear relationships. Optional models (XGBoost, LightGBM, CatBoost) appear only when installed.

Hyperparameter Tuning

Searches a parameter grid to find better model settings. Slower than a single fit but often improves predictive performance.

Search Strategy

Grid search evaluates every parameter combination; random search samples combinations and is faster on large grids.

Tuning CV Folds

Number of cross-validation folds used inside the hyperparameter search loop. More folds give stabler estimates at higher compute cost.

Pre-Modeling Statistics

Runs Pearson, Spearman, and ANOVA screens before modeling to summarise each feature's relationship with the target.

Impurity Importance

Mean decrease in impurity computed inside tree-based models. Fast and built-in, but biased toward features with many distinct values.

Permutation Importance

Shuffles each feature and measures the resulting accuracy drop, giving a model-agnostic estimate of feature contribution.

Categorical Encoding

How non-numeric columns are converted before saving or modeling: keep as-is, map to integer codes, or expand into one-hot dummy columns.

Encoding Model

How genotypes are converted to numeric values: additive counts effect alleles (0/1/2); dominant and recessive collapse genotypes by inheritance model.

Min MAF

Minimum Minor Allele Frequency. SNPs rarer than this threshold are filtered out as statistically uninformative (default 0.05).

Max SNP Missingness

Maximum fraction of missing genotype calls allowed per SNP before it is dropped from the analysis.

Max Sample Missingness

Maximum fraction of missing genotypes allowed per sample before the sample is excluded.

HWE Threshold

Hardy-Weinberg Equilibrium p-value cutoff. SNPs below it deviate from expected genotype frequencies and may indicate genotyping error.

LD Pruning

Removes highly correlated SNPs in linkage disequilibrium to reduce redundancy and multicollinearity before association testing.

LD r² Threshold

Pairwise correlation cutoff used during LD pruning. SNP pairs above this r² are reduced to a single representative.

Association Method

The regression framework for SNP-trait testing: linear regression for quantitative traits, logistic regression for binary case/control outcomes.

Genomic Control

Applies an inflation-factor correction to p-values to account for population stratification and reduce spurious associations.

Multiple Testing Correction

Controls false positives across many SNP tests. Bonferroni is strict; Benjamini-Hochberg FDR balances discovery rate and error control.

Significance Threshold

The genome-wide p-value cutoff for calling a SNP-trait association significant after multiple-testing correction (conventional: 5×10⁻⁸).

Join Column (Primary Key)

The shared variable used to align samples across different omics datasets. It must exist in all selected datasets with matching entry IDs.

Data Layout

How the omics blocks are organised: a single wide table splits columns by prefix; multiple files uses one file per block; transposed has samples in columns and features in rows.

Sample ID Column

The column identifying samples across blocks, used to align rows when joining multiple omics layers.

Block Rule

How columns are assigned to blocks in a single wide table: prefix splits on the first underscore (e.g. ENV_temp → ENV block); manual lets you assign columns explicitly.

Feature ID Column

In transposed layout, the column identifying features (rows). Never auto-detected — it must be specified explicitly.

Outcome Column

The categorical variable DIABLO discriminates. Required for DIABLO; classes with fewer than ~10 members trigger a warning.

Principal Components

Number of latent dimensions extracted during dimensionality reduction (PCA, CCA). More components capture more variance but may include noise.

Cluster Count

The expected number of biological or technical subgroups in the integrated data (e.g., for SNF or K-Means).

Latent Factors

For factor analytic models like MOFA2: the underlying biological drivers or technical variance components spanning multiple omics views.

DIABLO Components (ncomp)

Number of latent components for the multi-block PLS-DA model. More components capture more discriminative structure but risk overfitting.

DIABLO keepX

Number of features retained per block per component in the sparse PLS model. Use the same value for all blocks or specify per-block values.

DIABLO Design Matrix

Controls how blocks are linked: full links all blocks to each other; null treats blocks independently; custom allows pairwise weights.

DIABLO perf Folds

Number of cross-validation folds for the perf() error-rate estimation.

DIABLO perf Repeats

Number of times cross-validation is repeated for balanced error-rate (BER) stability.

DIABLO Random Seed

Integer seed for reproducible DIABLO results.

DIABLO Tuning

Opt-in tuning of keepX via tune.block.splsda. The estimated cost derives from grid_size × ncomp × nrepeat × folds.

MOFA2 Convergence Mode

Controls convergence criteria for the variational EM algorithm: fast stops early; slow runs more iterations for tighter convergence.

MOFA2 dropR2 Threshold

Minimum R-squared contribution for a factor to be retained. Factors explaining less variance than this threshold are pruned.

MOFA2 scale_views

Whether to scale each view to unit variance before fitting. Useful when views have very different magnitudes.

MOFA2 Random Seed

Integer seed for reproducible MOFA2 factor estimation.

Trained Model File

A pickled scikit-learn compatible model (.pkl, .pickle, .joblib) exported from a previous result. Works for both classification and regression. Only upload models you trust.

Identifier Column

Optional column used to label each row in the prediction output so results can be traced back to individual samples.

Univariate Screening

Runs Pearson, Spearman and ANOVA F tests of every feature against the target with FDR correction. Standalone screening produces a derived dataset containing only the significant features plus the target, ready for downstream preprocessing or modeling.

Value Column

Numeric column whose distribution is compared across groups by t-test, ANOVA, Mann-Whitney or Kruskal-Wallis.

Group Column

Categorical column defining comparison groups. t-test and Mann-Whitney require exactly two distinct values; ANOVA and Kruskal-Wallis accept two or more.

First Categorical Column

First variable in the chi-square test of independence; its distinct values form the contingency-table rows.

Second Categorical Column

Second variable in the chi-square test of independence; its distinct values form the contingency-table columns.

Significance Level (alpha)

P-value threshold for calling a test result statistically significant. 0.05 is standard; 0.01 is stricter.

Paired Samples

Switch the t-test to paired mode for matched observations (e.g. the same subjects measured before and after treatment).

Candidate Models

Regression algorithms benchmarked under identical cross-validation splits in a Model Comparison run.

Selection Threshold

Threshold for filter-type feature selection: minimum variance (Variance), maximum inter-feature correlation (Correlation), or FDR alpha (FDR).

Selector Estimator

Model whose internal feature weights drive SelectFromModel selection: Random Forest importances or Lasso coefficients.

Ordination Method

PCA on standardised data, PCoA (classical MDS), or NMDS on a chosen distance matrix — reduces a samples-by-features table to a few axes.

Distance Metric

Pairwise distance used by PCoA and NMDS: euclidean, braycurtis, manhattan or cosine.

Number of Axes

How many ordination axes to compute; two is enough for the standard scatter plot.

Feature Columns

Comma-separated numeric columns to ordinate; blank uses every numeric column.

Colour Points By

Optional column that colours the ordination scatter points, typically a grouping factor.

Clustering Method

K-Means partitions into k compact clusters; hierarchical agglomerative clustering merges rows bottom-up with a linkage rule.

Number of Clusters (k)

How many groups to split rows into; a small interpretable number works best for management zones.

Linkage Rule

How inter-cluster distance is measured in hierarchical clustering; Ward is the usual default.

Fixed Effect

Predictor of primary interest in a mixed model; categorical columns become factors automatically.

Predictor Columns

Independent variables in a GLM; categorical predictors are dummy-coded automatically.

GLM Family

Poisson for counts where mean ≈ variance; Negative Binomial for overdispersed counts.

NB Dispersion (alpha)

Negative-binomial dispersion parameter; larger alpha means more overdispersion.

Timepoint Columns

Numeric columns holding repeated measurements in wide format — one per timepoint, one row per subject.

Label-shuffling permutation count for permutation-based tests (PERMANOVA, ANOSIM, Mantel, IndVal). 999 is standard; larger values refine small p-values.

Numeric columns forming the second side of a Mantel test — measured environmental/trait variables whose pairwise distances are correlated with community distances.

Distance metric for the environmental side of a Mantel test (euclidean is the usual default).

Feature-engineering transform: arithmetic, log1p, polynomial interactions, longitudinal summaries, one-hot, or wide→long reshape.

Columns the operation reads; ratio and difference take exactly two.

Optional name for the created column; auto-generated when blank.

Highest power/interaction degree generated.

Emit only cross-products, skipping squared/power terms.

Prefix for longitudinal summary columns (_mean/_delta/_slope/_min/_max).

Identifier columns preserved per row in a wide→long reshape.

Latitude/northing coordinate per sample, used to build the spatial neighbour graph for Moran's I.

Longitude/easting coordinate per sample.

Nearest neighbours per point in the spatial weight matrix; 4–8 suits plot grids.

Between-Subject Factor

Optional grouping column tested on subject means alongside the within-subject time effect.

GWAS figure exports are browser-side only

The Manhattan and QQ plot export buttons (PNG, SVG, HTML, CSV) render the file in your browser with JavaScript. There is no server-side GWAS export endpoint, so GWAS figures are not included in combined /reports output and cannot be downloaded by API clients.

Pathway lookups can be slow on first run

GO/KEGG/Reactome enrichment queries external databases. A first run against a cold cache can occasionally time out at the web-server layer; resubmitting the same analysis normally succeeds immediately.

Group-column suggestions are heuristic

Group/category dropdowns list non-numeric columns plus integer-valued numeric columns with few distinct values. Low-cardinality count data (for example sparse species counts) can still appear — verify the column is a real grouping before running a test.

Ordination colouring depends on the column type

The ordination Colour by option maps a categorical column to a discrete legend and a numeric column to a continuous colourbar. Columns that are neither clearly categorical nor numeric may fall back to uncoloured points.