Skip to contents

Process proteomics data

Source: R/Processing.R
process_proteomics.Rd

Complete processing pipeline: filtering, normalization, imputation and differential expression analysis. Coordinates sub-modules Normalization.R, Imputation.R, and DEAnalysis.R.

Usage

process_proteomics(
  preprocessing,
  export_dir = NULL,
  min_reps_filter = NULL,
  min_groups_filter = 1,
  norm_method = "cycloess",
  cyclic_loess_method = "fast",
  cyclic_loess_iterations = 3,
  cyclic_loess_span = 0.7,
  batch_correct = FALSE,
  batch_column = "Batch",
  batch_algorithm = "ComBat",
  batch_ComBat_mode = 1,
  batch_covariates = NULL,
  batch_qualitycontrol = FALSE,
  imp_method = "combo",
  mar_method = "Impseqrob",
  mnar_method = "min",
  prop_na_mnar = 0.51,
  prop_present_mar = 0.5,
  min_present_mar = 1,
  require_n_conditions = 1,
  max_na_prop = NULL,
  method_args = list(),
  with_value = NA_real_,
  comparisons = NULL,
  control = NULL,
  logFC_threshold = 0,
  alpha = 0.05,
  eBayes_trend = NULL,
  eBayes_robust = NULL,
  de_method = "limma",
  covariate_df = NULL,
  covariate_column = NULL,
  bio_replicate_column = NULL,
  export_normalized = TRUE,
  export_imputed = TRUE,
  export_format = "tsv",
  export_volcano = TRUE,
  export_boxplot = TRUE,
  export_pca = TRUE,
  verbose = TRUE
)

Arguments

preprocessing

A proteomics_data object, as returned by preprocess_spectronaut(), preprocess_diann(), preprocess_tmt() or preprocess_lfq().

export_dir

Output directory for exported files. Defaults to NULL, which writes nothing to disk; pass a path to enable the exports controlled by export_normalized, export_imputed, export_volcano, export_boxplot and export_pca.

min_reps_filter

Minimum replicates for filtering. If NULL, auto-computed

min_groups_filter

Minimum groups for filtering (default: 1)

norm_method

Normalization method passed to normalize_proteomics() (default: "cycloess"). See normalize_proteomics() for all 13 options.

cyclic_loess_method

Cyclic Loess method: "fast" or "pairs" (default: "fast")

cyclic_loess_iterations

Number of iterations for Cyclic Loess (default: 3)

cyclic_loess_span

Span parameter for Cyclic Loess (default: 0.7)

batch_correct

Logical. Apply BERT batch correction after normalization (default: FALSE). Requires covariate_df with batch_column.

batch_column

Column in covariate_df containing batch assignments (default: "Batch"). Must have >= 2 unique values.

batch_algorithm

Batch correction algorithm: "ComBat" (default), "limma", or "ref"

batch_ComBat_mode

Integer 1-4 for ComBat parametric/mean-only settings (default: 1)

batch_covariates

Character vector of colData column names to protect during BERT batch correction (default: NULL). Categorical columns are encoded as indicator variables automatically; numeric columns are passed through unchanged and therefore act as continuous covariates.

batch_qualitycontrol

Logical. Compute ASW quality metrics (default: FALSE)

imp_method

Imputation method (default: "combo"). See impute_proteomics() for all options.

mar_method

MAR method for combo mode (default: "Impseqrob")

mnar_method

MNAR method for combo mode (default: "min")

prop_na_mnar

NA proportion for MNAR classification (default: 0.51)

prop_present_mar

Present proportion for MAR (default: 0.5)

min_present_mar

Minimum present values for MAR (default: 1)

require_n_conditions

Required conditions with presence (default: 1)

max_na_prop

Maximum NA proportion above which a protein is removed before imputation. NULL (the default) disables the filter.

method_args

Named list of per-method argument lists

with_value

Constant value for imp_method="with"

comparisons

Comparisons for DE. If NULL, generates all pairwise

control

Control condition. If NULL, compares all

logFC_threshold

LogFC threshold for significance (default: 0)

alpha

Adjusted p-value threshold (default: 0.05)

eBayes_trend

Use trend estimation in eBayes. If NULL (default), it is resolved from de_method: TRUE for "limma", FALSE for "limpa".

eBayes_robust

Use robust estimation in eBayes. If NULL (default), it is resolved from de_method: TRUE for "limma", FALSE for "limpa".

de_method

DE method: "limma" (default) or "limpa" (probabilistic, requires imp_method="limpa")

covariate_df

Data frame with Column + covariate column(s) for paired/blocked design (default: NULL)

covariate_column

Name(s) of the covariate column(s) for the DE model. Single string (e.g., "Subject") or character vector (e.g., c("Subject", "Batch")). Default: NULL

bio_replicate_column

Column name in covariate_df identifying biological replicates (e.g., "Patient", "Subject"). Used with limma::duplicateCorrelation() to account for technical replicates or paired designs via random effect blocking. Default: NULL

export_normalized

Export normalized matrix (default: TRUE)

export_imputed

Export imputed matrix (default: TRUE)

export_format

Export format: "tsv", "parquet", or "both" (default: "tsv")

export_volcano

Export VolcanoPlot_Input (default: TRUE)

export_boxplot

Export BoxPlot_Input (default: TRUE)

export_pca

Export PCA_Input (default: TRUE)

verbose

Print progress messages (default: TRUE)

Value

List of class proteomics_result with:

  • se_proc: Processed SummarizedExperiment with all assays

  • DEPs_results: Data frame with differential expression results

  • BoxPlot_Input: Long-format data frame with one row per protein, sample and assay (columns Column, Assay, Intensity, Condition, Replicate, Protein.IDs). Feed it to boxplot_highchart_list().

  • PCA_Input: Long-format data frame with the imputed intensities plus one adjP_* column per comparison and a sig_any flag. Feed it to pca_highchart_list(), proteomics_heatmap() or proteomics_heatmap_list().

  • comparisons: Comparisons performed

  • parameters: Parameters used

BoxPlot_Input and PCA_Input are always returned, whether or not export_dir is given: the plotting layer needs them, and requiring a round trip through disk to obtain them would be gratuitous.

Details

The input is a proteomics_data object, which every preprocessing function returns with the same three elements, so the pipeline is indifferent to the acquisition and the search engine the data came from.

Examples

data(nadia_dia)

# Normalization, imputation and differential expression in one call
res <- process_proteomics(nadia_dia, verbose = FALSE)
res
#> === Proteomics Processing Result ===
#> 
#> SummarizedExperiment:
#>   - Proteins: 1997 
#>   - Samples: 12 
#>   - Assays: raw, log2, cycloess, Impseqrob_min 
#>   - Conditions: A, B, D 
#> 
#> Differential Results:
#>   - Total rows: 5991 
#>   - Comparisons: B-A, D-A, D-B 
#>     B-A: Up=325, Down=172
#>     D-A: Up=374, Down=383
#>     D-B: Up=286, Down=382
#> 
#> Parameters:
#>   - Normalization: cycloess 
#>     - Cyclic Loess method: fast 
#>     - Cyclic Loess iterations: 3 
#>     - Cyclic Loess span: 0.7 
#>   - Imputation: combo 
#>     - MAR method: Impseqrob 
#>     - MNAR method: min 
#>   - Alpha: 0.05 
#>   - logFC threshold: 0 
#>   - Output directory: (no export) 

SummarizedExperiment::assayNames(res$se_proc)
#> [1] "raw"           "log2"          "cycloess"      "Impseqrob_min"
table(res$DEPs_results$Comparison, res$DEPs_results$Change)
#>      
#>         Up Down No Change
#>   B-A  325  172      1500
#>   D-A  374  383      1240
#>   D-B  286  382      1329

# The plotting layer is fed straight from the result, no files involved
names(res$BoxPlot_Input)
#> [1] "Column"      "Assay"       "Intensity"   "Condition"   "Replicate"  
#> [6] "Protein.IDs"
names(res$PCA_Input)
#> [1] "SampleID"  "FeatureID" "Intensity" "Condition" "Replicate" "adjP_B-A" 
#> [7] "adjP_D-A"  "adjP_D-B"  "sig_any"  

# Another combination: quantile normalization and a single imputation method
res2 <- process_proteomics(nadia_dia, norm_method = "quantile",
                           imp_method = "QRILC", verbose = FALSE)
SummarizedExperiment::assayNames(res2$se_proc)
#> [1] "raw"      "log2"     "quantile" "QRILC"