Dimension Reduction panel

Dimension Reduction panel

The Dimension Reduction panel on the left side allows users to configure the input data, feature selection strategy, and dimensionality reduction parameters.

1. Data for dimension reduction: This dropdown defines which data values are used for calculating feature variance and performing dimension reduction. Available options include:

   • Precalculated Dimension Reductions: Displays previously computed and saved dimension reduction results available in PanHunter.
   • Protein abundance / Expression values: Uses normalized abundance or expression values for variance calculation and embedding generation. Examples include:
     • Bulk RNA-Seq / Pseudobulk RNA-Seq: log-normalized counts (log-CPM)
     • ScreenSeq: log-normalized counts per 10,000 (log-CP10,000)
     • Proteomics: log-normalized protein abundances
   • Comparison analysis (logFC): Uses log fold-change values from selected comparison analyses.
   • Comparison analysis (PI-score): Uses PI-score values, which combine fold-change and statistical significance information.

2. Method: This dropdown specifies the dimensionality reduction algorithm used for embedding visualization. Available methods include: PCA,t-SNE,UMAP Refer to Dimension Reduction documentation for detailed information on the available dimension reduction options in PanHunter.

3. Select feature set: Defines how the feature subset for dimensionality reduction is selected. Available options include:

   • Custom Feature Lists: Allows selection of user-defined feature lists from the Feature List Selector table.

   

   • Feature List Collections: Uses predefined curated feature collections or pathway-based feature sets which can be selected from the Feature Collection Input dropdown. This dropdown allows selection of biological annotation databases or pathway collections used for feature filtering such as:
     • GO BP (Gene Ontology Biological Process)
     • GO CC (Cellular Component)
     • GO MF (Molecular Function)
     • MSigDB collections After selecting a collection category, a second dropdown allows selection of a specific feature collection or pathway term (e.g., a GO biological process).

   

4. Set number of features: Defines the maximum number of highly variable features used for dimension reduction. If the total number of available features exceeds the specified threshold, only the most variable features are retained for downstream analysis. The default value is 500

   

5. Select Principal Components: Specifies which principal components are displayed on the X- and Y-axes of the embedding plot. For example: PC1 vs PC2, PC2 vs PC3. The default values are:

   • X-axis: Principal Component 1
   • Y-axis: Principal Component 2

6. Set additional parameters: Additional method-specific parameters for dimensionality reduction can be configured in this section. Examples include:

   • number of principal components used for computation,
   • distance metrics,
   • initialization strategies,
   • preprocessing options. For PCA, t-SNE, and UMAP, this section also provides access to batch-effect correction and reshuffling options.

   

   1. Set number of PCs: Defines the total number of principal components to compute. The default value is 50.

      • PCA: determines the number of components available for visualization.
      • t-SNE / UMAP: principal components are used as input for downstream embedding calculations.

   2. Correct factors: Allows correction of unwanted technical or experimental effects (batch correction). Selected factors (e.g., replicate, batch, treatment) are adjusted to minimize systematic variation unrelated to biological differences.

      • After correction: group-specific average shifts are reduced, technical variability is minimized, sample comparability is improved. By default, factor correction is disabled.

   3. Random reshuffling: Provides reshuffling or permutation-based controls for assessing embedding robustness. Available reshuffling modes include Within sample and Within feature This option can be used to evaluate whether observed clustering patterns are biologically meaningful or arise from random structure in the dataset.

7. Save coordinates: Generated embedding coordinates can be saved within PanHunter using the Save in PanHunter option for later reuse and comparison.

8. Download feature list: The complete list of features used for dimensionality reduction can be downloaded using the Download feature list (XLSX) button.