Exploratory analysis panel

Exploratory Analysis Panel

The Exploratory Analysis panel provides tools for identifying metadata variables and molecular features that are associated with the currently selected dimensionality reduction or feature space. It is intended for exploratory hypothesis generation and helps uncover potential drivers of sample clustering, separation, or heterogeneity.

The analysis can be performed on:

• Sample annotations and categorical metadata variables
• Numerical sample metadata
• Molecular features such as genes, proteins, metabolites, or other measured entities

1. Coordinates to analyze setting defines the data space on which the exploratory analysis is performed. Available options include:

   

   • Current 2D plot uses the coordinates of the currently displayed dimensionality reduction plot (e.g., PCA, UMAP, t-SNE). It helps identify metadata variables that explain clustering observed in PCA or UMAP space.
   • Top features uses the currently selected subset of informative or highly variable features to focus the analysis while reducing noise.
   • All features uses the complete feature matrix available for the loaded samples to perform a comprehensive analysis using the full dataset.
     The selected coordinate space determines which sample relationships are investigated during the analysis.

2. NA as a class option determines how missing values in sample annotations are handled during categorical analyses. Available options are:

   

   • Yes: Missing values are treated as an independent category during analysis.
   • No: Samples with missing annotation values are excluded from the corresponding categorical analysis.

   Treating missing values as a separate class can be useful when the absence of information itself may be biologically or technically meaningful.

3. Minimum span parameter defines the minimum variability required for a feature to be included in exploratory feature analyses.

   

   The span is calculated as: > 90th percentile − 10th percentile for the feature values across all analyzed samples.

   Features with a span below the specified threshold are excluded from the analysis because they contain insufficient variation to contribute meaningful information.

   For example:

   Minimum span = 0.8
   

   retains only features whose values vary by at least 0.8 units between the 10th and 90th percentile.

   This filtering reduces noise originating from nearly constant features and improves computational efficiency.

4. Feature NA values to zero option controls how missing feature measurements are handled during exploratory feature analysis.

   

   • Enabled Missing feature values are replaced with zeros before analysis.
   • Disabled Missing values remain as missing and are ignored during calculations whenever possible.

   This option only affects feature-based analyses and does not influence analyses of sample metadata or annotations.

   Note: For transcriptomics data, enabling this option may approximate dropout handling in sparse datasets. For proteomics data, the appropriate setting depends on whether missing values are considered true absence of signal or missing measurements due to technical limitations.

5. Analysis modes: The panel provides three complementary exploratory analysis modes.

   

   • Analyze Categories function evaluates associations between the selected coordinate space and categorical sample annotations.

     Examples include:

     • Treatment groups
     • Tissue types
     • Disease states
     • Experimental batches
     • Patient cohorts

     This analysis helps identify which categorical variables best explain the observed sample organization.

     Typical questions addressed include:

     • Which metadata variable drives the PCA separation?
     • Are samples clustering according to treatment?
     • Is there evidence of batch effects?

   • Analyze Numerics function evaluates associations between the selected coordinate space and numerical metadata variables.

     Examples include:

     • Age
     • BMI
     • Sequencing depth
     • Cell count
     • Protein concentration
     • Quality control metrics

     This allows identification of continuous variables that correlate with major sources of variation in the dataset.

     Typical applications include:

     • Detecting technical confounders
     • Identifying quality-related trends
     • Exploring biological gradients

   • Analyze Features function identifies molecular features that contribute most strongly to the selected coordinate space. Depending on the data type, these features may include:

     • Genes
     • Proteins
     • Metabolites
     • Peaks
     • Other quantitative molecular measurements

     This analysis can help identify:

     • Marker genes or proteins driving sample separation
     • Features associated with biological subgroups
     • Candidate biomarkers for downstream investigation