Skip to content

08 Analysis

Jump to bottom
Charlie Bayne edited this page Mar 28, 2026 · 1 revision

Analysis

The Analysis tab is the core of Conduit-Summit, providing a full suite of tools for exploring and statistically analyzing your data. It is organized into nine sub-sections, accessible via tabs within the page.

The Aggregation Level header control determines what assay is analyzed across all sub-sections. The default is protein_groups, but you can switch to any available assay (domain, kingdom, phylum, class, order, family, genus, species, GO terms, KEGG pathways, peptides, precursors, etc.).

QC

Five sub-tabs providing essential quality control visualizations. These should be your first stop after loading data.

Feature Numbers

A stacked bar chart showing the number of features detected per sample. The color of each bar segment indicates how many samples share that feature — providing a quick view of feature overlap across your experiment.

Missing Values

A hierarchically clustered heatmap of missing values across the data matrix. Missing values are shown in one color; detected values in another. This reveals patterns in data completeness — for example, whether missing values cluster by sample group, which can indicate a biological or technical effect.

Column (sample) and row (feature) annotations can be added using any colData or rowData variable. Multiple annotation variables can be selected at once, though this can become visually complex.

Sample Correlation

A heatmap where each cell shows the Pearson correlation between two samples. The diagonal always shows a perfect correlation (1.0) since each sample is perfectly correlated with itself. High off-diagonal correlations within expected groups and lower correlations between groups indicate a well-performing experiment.

Sample-level metadata (colData) can be overlaid on the rows and columns to help identify which variables are driving the observed correlation structure.

Intensity Distribution

Two complementary plots visualizing the distribution of feature intensities:

  1. Density plot: Feature intensity distributions colored by whether the value was observed or imputed. In DIA data, missing values are typically due to low abundance, so imputed values should cluster at the low end of the intensity distribution.
  2. ECDF plot: Cumulative fraction of the data by log intensity. Shows what proportion of features are detected at or below each intensity threshold. Also colored by observed vs. imputed.

Density Plot

Displays the intensity distributions of each sample on a log scale as smoothed density curves. The plot options allow coloring by any sample-level variable (colData), making it easy to compare distributions across groups. This is useful for checking that normalization has worked correctly.

PCA

Principal Component Analysis (PCA) reduces the high-dimensional feature data to two dimensions (PC1 and PC2) that capture the most variation. The percentage of total variance explained by each PC is shown on the axis labels.

Options:

  • Color by: Choose any sample-level variable to color points
  • Shape by: Choose any sample-level variable to assign shapes (avoid variables with more than 6 levels)
  • Facet formula: Split the plot into panels by a variable (e.g., ~treatment). The default ~NULL shows all samples together. See the ggplot2 faceting docs for syntax.

Heatmap

A heatmap displaying features as rows and samples as columns. Two modes are available:

  • Static heatmap (pheatmap): Suitable for publication figures. Supports hierarchical clustering of rows and columns, and annotation tracks.
  • Interactive heatmap (plotly): Allows zooming and hovering over individual cells to read exact values. Useful for data exploration.

Options:

  • Number of top features: Limits the heatmap to the top N features by variance, which is strongly recommended for large assays. A heatmap with thousands of rows and hundreds of columns is difficult to interpret and slow to render.
  • Row annotations: Color rows by any feature-level variable (rowData)
  • Column annotations: Color columns by any sample-level variable (colData)

Relative Abundance

Stacked bar charts showing the relative compositional abundance of features (e.g., taxa) across samples. Useful for visualizing community composition at any taxonomic level.

Supports faceting to compare groups side-by-side.

Explore

A flexible custom plotting interface for ad hoc exploration of your data.

Options:

  • Group by: Summarize data by any colData variable
  • Summarization method: Mean, median, sum, standard deviation, or count
  • Plot type: Scatter, bar, line, boxplot, violin, or histogram
  • Aesthetics: X-axis, Y-axis, color, shape, and faceting

This tab is useful when you want a specific visualization not covered by the other analysis sub-tabs.

Statistics

Differential abundance analysis using the limma package, which applies linear models with empirical Bayes moderation. This approach is well-validated for proteomics data and is more statistically robust than simple t-tests, especially with smaller sample sizes.

If you are unfamiliar with general linear models, the StatQuest series on GLMs is an approachable introduction. For a proteomics-specific guide to limma, see this paper.

Formula

Enter the limma model formula. This describes the relationship between your features and the sample metadata. For example:

  • ~ treatment — model the effect of a single treatment variable
  • ~ treatment + batch — model treatment while accounting for a batch effect
  • ~ treatment * time — model treatment, time, and their interaction

Contrast

Specify which comparison to make. Contrasts define the numerator and denominator of your fold-change calculation. For example, treatmentKO - treatmentWT tests KO vs. WT.

Note on directionality: Features with positive log2 fold change are higher in the numerator condition; features with negative log2 fold change are higher in the denominator. The direction depends entirely on how you specify the contrast.

Statistics Table

A downloadable table of limma results for all features, including:

  • logFC: Log2 fold change
  • AveExpr: Average expression across all samples
  • t: Moderated t-statistic
  • P.Value: Raw p-value
  • adj.P.Val: Benjamini-Hochberg adjusted p-value
  • B: Log-odds of differential abundance

Volcano Plot

An interactive volcano plot with log2 fold change on the X-axis and -log10(adjusted p-value) on the Y-axis. Thresholds for fold change and significance can be adjusted in the sidebar.

Clicking on a point in the volcano plot opens a feature intensity plot showing that feature's values across all samples, colored by a user-selected metadata variable.

Enrichment Analysis

Gene set enrichment analysis to identify biological terms (GO, KEGG, or taxonomic groupings) that are systematically enriched among differentially abundant features.

Requires: Statistics results (limma must be run first).

Methods:

  • GSEA: Gene Set Enrichment Analysis — uses the full ranked list of features, no significance threshold required
  • ORA: Over-Representation Analysis — tests whether features above a fold-change threshold are enriched for a term

Term types:

  • GO (Gene Ontology)
  • KEGG pathways
  • Taxonomic levels (domain through species)

Visualizations: Dotplot, barplot, and cnetplot (network of enriched terms and associated features).

Results can be downloaded as a CSV.

Pathway Analysis

Visualizes KEGG pathway maps overlaid with your differential abundance statistics. Nodes on the pathway map are colored by log2 fold change, allowing you to see which parts of a metabolic or signaling pathway are most affected.

Annotation types:

  • KEGG Map Pathway (KO-based): Uses KEGG Orthology annotations
  • UniProt KEGG Pathway (organism-specific): Uses UniProt-to-KEGG mappings

An option to show only statistically significant features (above a user-defined fold change and p-value threshold) is available.

Requires: Statistics results (limma must be run first).

Classification / Prediction

Machine learning models to predict a sample-level outcome from the feature data. Useful for biomarker discovery and assessing how well the measured features can classify samples.

Options:

  • Outcome variable: Any colData variable (binary, multiclass, or continuous)
  • Model type: LASSO regression, Random Forest, or XGBoost
  • Train/test split: Configurable from 50–90% training data
  • Cross-validation folds: Adjustable

Outputs:

  • ROC curve (classification): Area under the curve (AUC) on the test set
  • Precision-recall curve (classification): Especially informative for imbalanced classes
  • Feature importance plot: Top N features ranked by their contribution to the model

Model training runs asynchronously in the background with a progress indicator, so you can continue exploring other tabs while it runs.

Clone this wiki locally