Metabolite abundance matrix filtering

metabolite-abundance-matrix-filtering

Summary

Filter metabolite abundance matrices by missing-value threshold and feature intersection across batches, then impute missing values using k-nearest neighbour. This skill reduces the feature space to metabolites reliably quantified across all batches while preserving signal integrity through principled imputation.

When to use

When you have multiple batches of metabolomics data in SummarizedExperiment format with log-transformed abundance assays and need to: (1) remove poorly-quantified metabolites with >50% missing values within each batch, (2) retain only metabolites present across all batches for cross-batch comparisons, and (3) impute remaining scattered missing values before downstream normalisation or statistical analysis.

When NOT to use

• Input data already has no missing values or has been pre-filtered to a single consistent feature set across batches — the filtering and imputation steps would be redundant.
• You require retention of metabolites present in only a subset of batches (e.g., for batch-specific biomarker discovery) — the intersect method discards batch-specific signals.
• Missing values are believed to be informative (e.g., true biological absence rather than measurement failure) — imputation by k-NN would introduce bias.

Inputs

• List of SummarizedExperiment objects (one per batch)
• Log-transformed metabolite abundance assay (e.g., 'logRaw' assay within each SummarizedExperiment)

Outputs

• List of SummarizedExperiment objects with new assay (e.g., 'rawImpute') containing filtered and imputed metabolite abundances
• Reduced metabolite feature set (metabolites retained across all batches)
• Complete abundance matrix with no missing values

How to apply

Load a list of SummarizedExperiment objects, each representing one batch with a log-transformed abundance assay (e.g., 'logRaw'). Apply hRUV::clean() with threshold=0.5, method='intersect', specifying the source assay and a new assay name (e.g., 'rawImpute'). The function first filters out metabolites with ≥50% missing values per batch, then retains only those quantified (non-missing) in all batches via set intersection, and finally performs k-nearest neighbour imputation on the remaining missing values. Verify the output by confirming metabolite count reduction, zero missing values in the new assay, and consistent dimensions across all batches.

Related tools

• hRUV (Applies clean() function to filter metabolites by missing-value threshold and intersect method, then performs k-NN imputation on the filtered abundance matrix.) — https://github.com/SydneyBioX/hRUV
• SummarizedExperiment (Data container for storing and accessing batch-wise metabolite abundance assays with consistent row/column structure.)
• DMwR2 (Provides k-nearest neighbour imputation algorithm used by hRUV::clean() to fill remaining missing values.)
• dplyr (Supporting utility for data manipulation and verification of filtering results.)

Examples

dat_list = hRUV::clean(dat_list, threshold = 0.5, method = "intersect", assay = "logRaw", newAssay = "rawImpute")

Evaluation signals

• Metabolite count in output assay is smaller than input (confirms filtering occurred) and matches the number of features quantified in all batches.
• Output assay contains zero missing values (NA or NaN counts are 0).
• Dimensions (samples × metabolites) are identical across all batches in the output list.
• Metabolites retained are a subset of the input metabolites; no novel metabolites appear in the output.
• Log-transformed abundance values in the imputed assay remain within plausible range (no extreme outliers introduced by k-NN imputation).

Limitations

• The 50% missing-value threshold is fixed in the workflow; if data has a different missingness pattern, the threshold may need adjustment (not explored in the article).
• The intersect method is conservative — metabolites unique to any single batch are discarded, which may lose batch-specific signals or rare metabolites.
• k-NN imputation relies on abundance similarity among samples; it may perform poorly in low-sample-count batches or when metabolite abundance distribution is non-metric.
• No changelog provided in the hRUV repository, limiting traceability of parameter defaults or imputation algorithm versions.

Evidence

• [intro] The clean function filters metabolites with >50% missing values per batch using the intersect method, then performs k-nearest neighbour imputation to produce a rawImpute assay containing 61 metabolites quantified across all five batches.: "The clean function filters metabolites with >50% missing values per batch using the intersect method, then performs k-nearest neighbour imputation"
• [intro] We have filtered metabolites with more than 50% of missing values per batch and selected metabolites that are quantified across all batches (intersect): "We have filtered metabolites with more than 50% of missing values per batch and selected metabolites that are quantified across all batches (intersect)"
• [intro] dat_list = hRUV::clean(dat_list, threshold = 0.5, method = "intersect", assay = "logRaw", newAssay = "rawImpute"): "dat_list = hRUV::clean(dat_list, threshold = 0.5, method = "intersect", assay = "logRaw", newAssay = "rawImpute")"
• [readme] hRUV is a package for normalisation of multiple batches of metabolomics data in a hierarchical strategy: "hRUV is a package for normalisation of multiple batches of metabolomics data in a hierarchical strategy"
• [intro] assay(dat, "logRaw", withDimnames = FALSE) = log2(assay(dat, "raw") + 1): "assay(dat, "logRaw", withDimnames = FALSE) = log2(assay(dat, "raw") + 1)"