Installation

NADIA can be installed with BiocManager. The heatmap and Pattern Profiler packages are optional dependencies and are installed separately when those visualisations are required.

if (!requireNamespace("BiocManager", quietly = TRUE))
  install.packages("BiocManager")

BiocManager::install(c("NADIA", "ComplexHeatmap", "Mfuzz", "Biobase"))
install.packages(c("tidyHeatmap", "circlize", "e1071"))

Processing the example dataset

The nadia_dia dataset contains three conditions (A, B, and D) with four replicates per condition. process_proteomics() returns the processed experiment and the long-format tables expected by the plotting and clustering functions.

library(NADIA)
data(nadia_dia)

res <- process_proteomics(nadia_dia, verbose = FALSE)

The sample design should always be checked before a plot is interpreted.

sample_design <- unique(
  res$PCA_Input[c("SampleID", "Condition", "Replicate")]
)

design_summary <- as.data.frame(table(sample_design$Condition))
names(design_summary) <- c("Condition", "Samples")

knitr::kable(
  design_summary,
  row.names = FALSE,
  caption = "Number of samples in each condition of the example experiment."
)

Most plotting functions in this vignette receive ordinary data frames, such as DEPs_results, BoxPlot_Input, or PCA_Input. Compatible tables produced by another workflow can therefore be plotted without recreating the original SummarizedExperiment.

Pattern Profiler separates clustering from visualisation. pattern_profiler_analysis() uses se_proc and DEPs_results to produce long_output; the plotting functions use this data frame directly, so saved clustering results can be visualised without rerunning the analysis.

input_summary <- data.frame(
  Object = c("DEPs_results", "BoxPlot_Input", "PCA_Input", "se_proc"),
  Rows = c(
    nrow(res$DEPs_results),
    nrow(res$BoxPlot_Input),
    nrow(res$PCA_Input),
    nrow(res$se_proc)
  ),
  Columns = c(
    ncol(res$DEPs_results),
    ncol(res$BoxPlot_Input),
    ncol(res$PCA_Input),
    ncol(res$se_proc)
  ),
  Used_by = c(
    "Volcano plots and Pattern Profiler",
    "Boxplots",
    "PCA and heatmaps",
    "Pattern Profiler"
  )
)

knitr::kable(
  input_summary,
  row.names = FALSE,
  caption = "Objects returned by process_proteomics() for visualisation."
)

The values are calculated directly with nrow() and ncol(), but the row count reflects the structure of each object rather than the number of unique proteins. DEPs_results contains 1,997 proteins across three comparisons (5,991 rows), BoxPlot_Input contains the same proteins across 12 samples and two assays (47,928 rows), and PCA_Input contains one row per protein and sample (23,964 rows). In the three data frames, Columns counts the available variables; in se_proc, its 1,997 rows and 12 columns correspond directly to proteins and samples, respectively.

Interactive and static output

Interactive charts are particularly useful during exploration: points can be identified from tooltips, axes can be zoomed, and groups can be hidden from the legend. A static heatmap is preferable when dimensions, typography, and vector output must remain fixed for a manuscript.

One volcano plot, one boxplot, one PCA plot, and two compact Pattern Profiler charts are rendered below. Additional variants are shown as code without execution to keep the installed vignette at a reasonable size. The download module is omitted from these five embedded widgets; charts created in a regular R session retain their export menu.

Highcharts licence

Summarising the selected comparison

Before drawing the figure, it is helpful to count the proteins assigned to each category. The table below applies the same thresholds as the plot: adjusted p-value below 0.05 and absolute log2 fold change of at least 1.

de_ba <- res$DEPs_results[
  res$DEPs_results$Comparison == "B-A",
  ,
  drop = FALSE
]

volcano_class <- ifelse(
  !is.na(de_ba$adj.P.Val) & de_ba$adj.P.Val < 0.05 & de_ba$logFC >= 1,
  "Up",
  ifelse(
    !is.na(de_ba$adj.P.Val) & de_ba$adj.P.Val < 0.05 & de_ba$logFC <= -1,
    "Down",
    "Not significant"
  )
)

volcano_counts <- as.data.frame(table(
  factor(volcano_class, levels = c("Up", "Down", "Not significant"))
))
names(volcano_counts) <- c("Classification", "Proteins")

n_up <- volcano_counts$Proteins[volcano_counts$Classification == "Up"]
n_down <- volcano_counts$Proteins[volcano_counts$Classification == "Down"]
n_not_significant <- volcano_counts$Proteins[
  volcano_counts$Classification == "Not significant"
]

knitr::kable(
  volcano_counts,
  row.names = FALSE,
  caption = "Protein classification for B-A at adjusted p < 0.05 and |log2FC| >= 1."
)

Under these criteria, the example contains 264 up-regulated proteins, 10 down-regulated proteins, and 1723 proteins that do not pass both thresholds. The table provides exact counts, which are difficult to estimate from the volcano plot because many points overlap.

Creating and reading the plot

volcano_highchart_list() creates one chart for every requested comparison. Here, the six most statistically significant proteins that also pass both thresholds are labelled automatically.

volcanoes <- volcano_highchart_list(
  res$DEPs_results,
  comparisons = "B-A",
  alpha = 0.05,
  lfc_thr = 1,
  p_col = "adj.P.Val",
  show_top_genes = 6,
  title = "Differential abundance: {comparison}"
)

prepare_vignette_highchart(volcanoes[["B-A"]])

The x-axis is the log2 fold change. Positive values indicate greater abundance in B relative to A, whereas negative values indicate greater abundance in A. The y-axis is -log10(adj.P.Val), so stronger statistical evidence appears higher in the chart. A protein is coloured as up- or down-regulated only when it passes both the horizontal significance threshold and the vertical effect-size threshold.

The lfc_thr argument controls the guides and classification in the figure. It does not rerun differential-abundance analysis or alter res$DEPs_results. For the plot and the result table to communicate the same decision rule, use the same alpha and fold-change threshold in both stages and record those values.

By default, p_col = "adj.P.Val" controls both the y-axis and classification. Raw P.Value can be displayed for diagnostic work, but adjusted values are the appropriate default when many proteins are tested simultaneously.

Labelling selected proteins and multiple comparisons

Automatic labels are convenient for exploration, while a manually selected set is usually clearer in a report. highlight_genes matches the Gene.Names column and may highlight a protein whether or not it passes the thresholds.

# Label proteins selected for their biological relevance.
highlighted_volcanoes <- volcano_highchart_list(
  res$DEPs_results,
  comparisons = "B-A",
  alpha = 0.05,
  lfc_thr = 1,
  highlight_genes = c("lamB", "yciU", "gltA")
)

highlighted_volcanoes[["B-A"]]

# Build several charts with identical settings.
volcanoes_all <- volcano_highchart_list(
  res$DEPs_results,
  comparisons = c("B-A", "D-A", "D-B"),
  alpha = 0.05,
  lfc_thr = 1,
  title = "Differential abundance: {comparison}"
)

names(volcanoes_all)

volcanoes_all[["B-A"]]
volcanoes_all[["D-A"]]
volcanoes_all[["D-B"]]

Comparing assays numerically

The example output contains the initial log2 intensities and the final processed assay. The following summary calculates the median for every sample and then reports the range of those sample medians within each assay.

sample_medians <- aggregate(
  Intensity ~ Assay + Column + Condition,
  data = res$BoxPlot_Input,
  FUN = median,
  na.rm = TRUE
)

median_ranges <- do.call(
  rbind,
  lapply(split(sample_medians$Intensity, sample_medians$Assay), function(x) {
    data.frame(
      Minimum = min(x),
      Maximum = max(x),
      Range = max(x) - min(x)
    )
  })
)

median_ranges$Assay <- rownames(median_ranges)
rownames(median_ranges) <- NULL
median_ranges <- median_ranges[c("Assay", "Minimum", "Maximum", "Range")]
median_ranges[-1] <- lapply(median_ranges[-1], round, digits = 3)

knitr::kable(
  median_ranges,
  row.names = FALSE,
  caption = "Range of sample medians before and after processing."
)

In this dataset, the range of sample medians decreases after processing. This is consistent with improved alignment of the sample distributions, but it is not proof that every technical effect has been removed. Distributional similarity should be considered together with the experimental design, PCA, and the normalisation diagnostics described in vignette("choosing-methods").

Creating and reading the boxplot

boxplot_highchart_list() creates one chart for every requested assay. Here, the list contains only the final processed assay. Each box spans the first to the third quartile, the central line is the median, and the whiskers extend to the most extreme values within 1.5 interquartile ranges. Hovering over a box reports these statistics and the number of finite protein intensities in that sample.

set.seed(42)

boxplots <- boxplot_highchart_list(
  res$BoxPlot_Input,
  assays = "Impseqrob_min",
  color_by = "Condition",
  group_order = c("A", "B", "D"),
  horizontal = FALSE,
  show_outliers = TRUE,
  title = "Processed intensity distributions",
  subtitle = "Assay: {assay}",
  height = 560
)

prepare_vignette_highchart(boxplots[["Impseqrob_min"]])

The medians are closely aligned, while each sample retains a broad intensity range. Alignment is expected after the processing used in this example; complete collapse of the boxes, however, would be suspicious because biological and sampling variation should remain.

The points beyond the whiskers are displayed with show_outliers = TRUE. They highlight the tails of each sample distribution but are not automatically erroneous measurements. In much larger experiments, set show_outliers = FALSE when the additional points obscure the boxes or make the interactive widget unnecessarily heavy.

# Display both stages as separate list elements.
boxplots_both <- boxplot_highchart_list(
  res$BoxPlot_Input,
  assays = c("log2", "Impseqrob_min"),
  group_order = c("A", "B", "D"),
  show_outliers = TRUE,
  outlier_size = 2,
  box_width = 16
)

boxplots_both[["log2"]]
boxplots_both[["Impseqrob_min"]]

Comparing protein subsets

The proteins included in a PCA influence the pattern of sample variation and separation that can be observed. NADIA supports all proteins, proteins significant in any comparison, or proteins significant in one named comparison. The table below compares the first two choices using the same centering and scaling settings.

pca_scores_all <- build_pca_scores(
  res$PCA_Input,
  mode = "all",
  center = TRUE,
  scale. = TRUE
)

pca_scores_any <- build_pca_scores(
  res$PCA_Input,
  mode = "any",
  center = TRUE,
  scale. = TRUE
)

pca_summary <- data.frame(
  Subset = c("All proteins", "Significant in any comparison"),
  Proteins = c(
    length(unique(res$PCA_Input$FeatureID)),
    length(unique(res$PCA_Input$FeatureID[res$PCA_Input$sig_any %in% TRUE]))
  ),
  PC1 = c(unique(pca_scores_all$PC1_Perc), unique(pca_scores_any$PC1_Perc)),
  PC2 = c(unique(pca_scores_all$PC2_Perc), unique(pca_scores_any$PC2_Perc))
)

names(pca_summary)[3:4] <- c("PC1 variance (%)", "PC2 variance (%)")

knitr::kable(
  pca_summary,
  row.names = FALSE,
  digits = 2,
  caption = "Protein subsets and variance represented by the first two principal components."
)

With all proteins, PC1 explains 37.59% of the variance. Restricting the analysis to proteins already selected by differential abundance raises this to 79.26%. The stronger separation in the selected subset is expected because the same data were used to select proteins that differ between conditions. It should not be presented as independent validation of those differences.

Creating and reading the PCA plot

pca_highchart_list() can generate several PCA plots in a single call. The resulting plots are stored in a named list, with one element for each protein subset requested through modes. In this example, modes = "all" includes every protein and therefore produces a single baseline PCA plot. Sample names are displayed because the dataset contains only 12 samples. The 95% confidence ellipses summarise where the samples from each condition are concentrated and how they are distributed in the PCA space.

condition_colours <- c(
  A = "#0072B2",
  B = "#D55E00",
  D = "#009E73"
)

pcas <- pca_highchart_list(
  res$PCA_Input,
  modes = "all",
  color_by = "Condition",
  group_order = c("A", "B", "D"),
  palette = condition_colours,
  addEllipses = TRUE,
  ellipse_type = "confidence",
  ellipse_level = 0.95,
  show_labels = TRUE,
  point_size = 5,
  center = TRUE,
  scale. = TRUE
)

prepare_vignette_highchart(pcas[["all"]])

Nearby points represent samples with similar multivariate protein profiles. Separation highlights major variation, but its biological or technical origin should be checked against the study design and sample-level quality metrics.

The 95% ellipses use each condition’s mean and covariance matrix together with a two-dimensional chi-squared contour to summarise group centre, spread, and orientation. They support visual comparison, while formal separation can be assessed with an appropriate statistical test.

With scale. = TRUE, proteins are standardised to unit variance and contribute more comparably to the PCA. With scale. = FALSE, more variable proteins have greater influence. Because scaling can change sample positions and apparent group separation, the selected setting should be reported.

Subset modes, ellipses, and sample filtering

To focus the PCA on proteins that are significant in a particular comparison, pass its name to modes. For example, modes = "B-A" selects the proteins that differ significantly between conditions B and A. The filter_samples_to_comparison argument then determines which samples are shown: TRUE keeps only samples from A and B, whereas FALSE plots all samples, including those from condition D, using the same set of proteins selected from the B-A comparison.

pca_ba <- pca_highchart_list(
  res$PCA_Input,
  modes = "B-A",
  alpha = 0.05,
  filter_samples_to_comparison = TRUE,
  group_order = c("A", "B"),
  palette = condition_colours[c("A", "B")],
  addEllipses = FALSE,
  show_labels = TRUE
)

pca_ba[["B-A"]]

At least three samples are required to draw either outline. With ellipse_type = "confidence" and ellipse_level = 0.95, the ellipse combines each group’s mean and covariance matrix with a two-dimensional chi-squared contour. It therefore provides a consistent summary of group centre, spread, and orientation, while the displayed points show the individual samples on which the outline is based.

Selecting proteins for the heatmap

Showing every quantified protein is possible but rarely legible. NADIA provides three filtering modes, and feature_ids can be used when a focused, explicitly defined panel is needed.

n_all <- length(unique(res$PCA_Input$FeatureID))
n_any <- length(unique(
  res$PCA_Input$FeatureID[res$PCA_Input$sig_any %in% TRUE]
))
n_ba <- length(unique(
  res$PCA_Input$FeatureID[
    !is.na(res$PCA_Input[["adjP_B-A"]]) &
      res$PCA_Input[["adjP_B-A"]] < 0.05
  ]
))

heatmap_subsets <- data.frame(
  Request = c(
    'mode = "all"',
    'mode = "any"',
    'mode = "target", comparison = "B-A"'
  ),
  Proteins = c(n_all, n_any, n_ba),
  Samples = c(12, 12, 8),
  Interpretation = c(
    "All quantified proteins",
    "Significant in at least one comparison",
    "Significant in B-A; only A and B samples"
  )
)

knitr::kable(
  heatmap_subsets,
  row.names = FALSE,
  caption = "Available protein-filtering modes for the example heatmap."
)

Unlike a comparison-specific PCA, mode = "target" in proteomics_heatmap() automatically restricts the samples to the conditions in the comparison. The B-A heatmap therefore contains eight samples rather than all twelve.

Creating and reading a focused heatmap

The complete B-A set contains 497 proteins, which is too dense for row labels in a vignette. The following code selects the 30 proteins with the smallest adjusted p-values among those passing the 0.05 threshold.

de_ba_ordered <- de_ba[
  !is.na(de_ba$adj.P.Val) & de_ba$adj.P.Val < 0.05,
  ,
  drop = FALSE
]
de_ba_ordered <- de_ba_ordered[order(de_ba_ordered$adj.P.Val), , drop = FALSE]
ba_top_ids <- head(unique(de_ba_ordered$Protein.IDs), 30)

proteomics_heatmap() creates one heatmap for the requested protein and sample subset. The selected IDs are supplied through feature_ids, while mode = "target" retains the samples from conditions A and B in the B-A comparison.

proteomics_heatmap(
  res$PCA_Input,
  mode = "target",
  comparison = "B-A",
  feature_ids = ba_top_ids,
  scale_data = "row",
  sample_order = "clustering",
  condition_order = c("A", "B"),
  cluster_rows = TRUE,
  cluster_columns = TRUE,
  show_row_names = TRUE,
  palette_value = "brewer:RdBu",
  reverse_palette = TRUE,
  heatmap_title = "Top B-A protein profiles"
)

Row-scaled abundance profiles for the 30 proteins with the smallest adjusted p-values in the B-A comparison.

Each row is transformed to a z-score because scale_data = "row". Warm and cool colours therefore represent values above and below that protein’s own mean; they do not indicate that one protein is more abundant than another. Row clustering groups proteins with similar relative profiles, while column clustering places samples with similar profiles next to one another. The condition annotation can then be used to assess whether the resulting sample clusters are consistent with conditions A and B.

Use scale_data = "none" when absolute log2 intensity differences between proteins are the subject of the figure. In that case, a small number of high-abundance proteins may dominate the colour range. Column scaling answers a different question and is rarely the default for protein-profile heatmaps.

Annotations and repeated heatmaps

Categorical protein annotations can be supplied as a data frame containing a FeatureID column. Rows may be split by one annotation and ordered by another. Because the 30 proteins used above all have positive logFC values, they would produce only one annotation category. For a clearer demonstration, the example below selects the 15 most significant proteins with higher abundance in B and the 15 most significant proteins with higher abundance in A. The resulting annotation and row split are displayed beneath the code.

ba_higher_b_ids <- head(unique(
  de_ba_ordered$Protein.IDs[de_ba_ordered$logFC > 0]
), 15)

ba_higher_a_ids <- head(unique(
  de_ba_ordered$Protein.IDs[de_ba_ordered$logFC < 0]
), 15)

ba_annotation_ids <- c(ba_higher_b_ids, ba_higher_a_ids)

direction <- ifelse(
  de_ba$logFC[match(ba_annotation_ids, de_ba$Protein.IDs)] > 0,
  "Higher in B",
  "Higher in A"
)

protein_annotation <- data.frame(
  FeatureID = ba_annotation_ids,
  Direction = direction
)

annotated_heatmap <- proteomics_heatmap(
  res$PCA_Input,
  mode = "target",
  comparison = "B-A",
  feature_ids = ba_annotation_ids,
  row_annotation = protein_annotation,
  row_annotation_cols = "Direction",
  row_annotation_palette = list(
    Direction = c("Higher in B" = "#D55E00", "Higher in A" = "#0072B2")
  ),
  split_rows_by = "Direction",
  scale_data = "row",
  sample_order = "condition",
  condition_order = c("A", "B")
)

Heatmap of 15 significant proteins with higher abundance in A and 15 with higher abundance in B, annotated and split by direction.

The Direction annotation is derived from the sign of logFC: positive values are labelled Higher in B, whereas negative values are labelled Higher in A. The coloured row annotation identifies these categories, and split_rows_by = "Direction" separates them into two heatmap sections. Within each section, the row-scaled colours still represent values above or below each protein’s own mean. This balanced selection is used only to illustrate the annotation and should not be interpreted as the observed proportion of up- and down-regulated proteins in the complete comparison.

# Build a named list containing several heatmaps.
heatmaps <- proteomics_heatmap_list(
  res$PCA_Input,
  modes = c("any", "B-A", "D-A"),
  scale_data = "row",
  sample_order = "condition",
  condition_order = c("A", "B", "D")
)

# Display the heatmaps individually.
heatmaps[["any"]]
heatmaps[["B-A"]]
heatmaps[["D-A"]]

proteomics_heatmap_list() returns a named list, so indexing it by mode displays each heatmap separately.

Building and summarising the example clusters

For a concise and reproducible visual example, the analysis below fixes the number of clusters at two. In a new study, the candidate solutions should first be compared with auto_select_c = TRUE and interpreted using the diagnostics described in the dedicated Pattern Profiler vignette.

pp_vis <- pattern_profiler_analysis(
  se_proc = res$se_proc,
  DEPs_results = res$DEPs_results,
  assay_name = "Impseqrob_min",
  filter_mode = "any",
  alpha = 0.05,
  condition_order = c("A", "B", "D"),
  aggregate = "median",
  auto_select_c = FALSE,
  c = 2,
  min_membership = 0.25,
  seed = 42,
  verbose = FALSE
)

A stricter membership threshold is useful for displaying the clearest examples of each pattern. The table reports how many proteins remain at 0.70 and the median standardised profile represented by each cluster.

display_membership <- 0.70
pp_display <- pp_vis$long_output[
  pp_vis$long_output$Membership >= display_membership,
  ,
  drop = FALSE
]

pattern_summary <- do.call(
  rbind,
  lapply(split(pp_display, pp_display$Cluster), function(x) {
    data.frame(
      Cluster = paste("Cluster", unique(x$Cluster)),
      Proteins = length(unique(x$FeatureID)),
      `Median membership` = median(x$Membership),
      A = median(x$A),
      B = median(x$B),
      D = median(x$D),
      check.names = FALSE
    )
  })
)

pattern_summary[c("Median membership", "A", "B", "D")] <- lapply(
  pattern_summary[c("Median membership", "A", "B", "D")],
  round,
  digits = 3
)

knitr::kable(
  pattern_summary,
  row.names = FALSE,
  caption = paste0(
    "High-membership proteins and median z-score profiles at membership >= ",
    display_membership, "."
  )
)

The two clusters describe opposing relative trajectories. One pattern remains comparatively high in A and B before decreasing in D, whereas the other rises across the ordered conditions. Because the condition values are within-protein z-scores, they describe profile shape rather than absolute protein abundance.

Comparing cluster centroids

cluster_centroids_highchart() creates one interactive chart containing the representative trajectories of all selected clusters. The median is used here because it is less sensitive than the mean to proteins near a cluster boundary.

centroid_plot <- cluster_centroids_highchart(
  pp_vis$long_output,
  conditions = pp_vis$conditions,
  min_membership = display_membership,
  centroid_summary = "median",
  title = "Pattern Profiler cluster centroids",
  height = 440
)

prepare_vignette_highchart(centroid_plot)

The separation between the centroid lines shows when the dominant patterns diverge. Hovering over a marker reports the condition and median z-score. The centroids summarise many proteins and should therefore be read together with the cluster sizes and memberships in the preceding table.

Inspecting representative protein profiles

cluster_profile_highchart() creates one interactive chart for a selected cluster, allowing individual protein profiles to be compared with their centroid. This reveals whether the profiles are concentrated around the summary or show substantial heterogeneity. To keep the interactive vignette responsive, the next chart displays 100 proteins sampled evenly across the membership range retained for cluster 1.

cluster_one <- pp_display[pp_display$Cluster == 1, , drop = FALSE]
cluster_one <- cluster_one[order(cluster_one$Membership, decreasing = TRUE), ]
representative_positions <- unique(round(seq(
  from = 1,
  to = nrow(cluster_one),
  length.out = min(100, nrow(cluster_one))
)))
representative_ids <- cluster_one$FeatureID[representative_positions]
representative_profiles <- cluster_one[
  cluster_one$FeatureID %in% representative_ids,
  ,
  drop = FALSE
]

profile_plot <- cluster_profile_highchart(
  representative_profiles,
  cluster = 1,
  conditions = pp_vis$conditions,
  min_membership = display_membership,
  centroid_summary = "median",
  line_opacity = 0.35,
  title = "Cluster 1: 100 representative protein profiles",
  height = 440
)

prepare_vignette_highchart(profile_plot)

The thin lines are individual standardised protein profiles and the thicker line is their median. A narrow band around the centroid indicates a coherent pattern; marked deviations identify proteins whose assignment deserves closer inspection. Showing only high-membership representatives makes the dominant shape clear, but it does not describe the full uncertainty of the fuzzy clustering.

The complete interactive list can be generated when detailed exploration is more important than vignette size. With hundreds of series, these charts are best inspected in an R session rather than embedded in the installed vignette.

all_profile_plots <- cluster_profile_highchart_list(
  pp_vis$long_output,
  conditions = pp_vis$conditions,
  min_membership = 0.70,
  centroid_summary = "median"
)

all_profile_plots[["Cluster_1"]]

Interactive charts

Every NADIA Highchart includes a download menu for PNG, SVG, and PDF output. The menu is disabled only in the five embedded examples above to keep this self-contained vignette compact and robust. For a fully interactive copy, save the original chart or selected list element as an HTML widget.

htmlwidgets::saveWidget(
  volcanoes[["B-A"]],
  file = file.path(tempdir(), "volcano_B-A.html"),
  selfcontained = TRUE
)

htmlwidgets::saveWidget(
  centroid_plot,
  file = file.path(tempdir(), "pattern_centroids.html"),
  selfcontained = TRUE
)

A self-contained widget is convenient for sharing, but it remains subject to the Highcharts licence described above. The file also contains the plotted data, so it should not be used to distribute confidential results without review.

Static heatmaps

For proteomics_heatmap(), export_path writes the displayed matrix and metadata to a TSV file, whereas export_file writes the figure. The plot format is inferred from the extension (.png, .svg, or .pdf).

proteomics_heatmap(
  res$PCA_Input,
  mode = "target",
  comparison = "B-A",
  feature_ids = ba_top_ids,
  scale_data = "row",
  sample_order = "condition",
  condition_order = c("A", "B"),
  export_path = file.path(tempdir(), "heatmap_B-A_data.tsv"),
  export_file = file.path(tempdir(), "heatmap_B-A.png"),
  plot_width = 8,
  plot_height = 7,
  export_dpi = 300
)

When proteomics_heatmap_list() is used, the mode is appended to each export name. For example, a base name of heatmap.png produces files such as heatmap_any.png and heatmap_B-A.png.

The companion vignette vignette("results-and-export") describes additional ways to save complete NADIA analyses, result tables, and interactive objects.