metabolomics-batch-effect-correction
Summary
Apply hierarchical RUV-III normalisation with intra-batch loess smoothing and inter-batch concatenate structure to remove batch effects and signal drift from multi-batch metabolomics data. This skill leverages sample replicates embedded within and across experimental runs to estimate and correct unwanted variation in SummarizedExperiment objects.
When to use
Apply this skill when you have log-transformed, imputed metabolomics data from multiple experimental batches (rawImpute assay) that shows visible batch effects in PCA plots or signal drift across runs, and your study design includes intra-batch replicates (technical replicates within a run) and inter-batch replicates (quality control or spike-in samples across batches) that can inform RUV-III decomposition.
When NOT to use
- Input data lacks intra-batch or inter-batch replicate samples; RUV-III requires both types to estimate unwanted variation.
- Metabolite filtering has not been performed; apply threshold filtering (>50% missing per batch) and intersect filtering before calling hRUV.
- Input assay is not log-transformed and imputed; hRUV expects rawImpute (log2-transformed and k-NN imputed) as input, not raw counts or unimputed data.
Inputs
- SummarizedExperiment object with rawImpute assay (log-transformed and k-NN imputed)
- Batch metadata (batch assignment per sample)
- Replicate metadata (intra-batch replicate identifiers; inter-batch replicate identifiers)
Outputs
- SummarizedExperiment object with new loessShort_concatenate assay (batch-corrected, normalised metabolite intensities)
- Batch effect correction applied to all metabolites quantified across all batches (intersect set)
How to apply
Load a SummarizedExperiment object containing the rawImpute assay and batch/replicate metadata. Call the hRUV function with parameters: intra='loessShort' to apply non-linear smoothing within batches using RUV-III with k=5 short replicate neighbors, and inter='concatenate' to apply hierarchical inter-batch normalisation using batch replicate samples with k=5 batch replicate neighbors. Extract the resulting normalised loessShort_concatenate assay from the returned SummarizedExperiment. Evaluate success by verifying that batch structure is no longer visible in PCA plots and that signal drift across experimental runs has been corrected.
Related tools
- hRUV (Core function implementing hierarchical RUV-III normalisation with intra-batch loess smoothing and inter-batch concatenate structure) — https://github.com/SydneyBioX/hRUV
- RUV-III (Statistical method underlying hRUV for estimating and removing unwanted variation using intra-batch and inter-batch replicate samples)
- SummarizedExperiment (Data container for storing assays, metadata, and batch/replicate annotations)
- dplyr (Tidyverse package for data manipulation and preparation of metadata)
Examples
library(hRUV); dat_normalized <- hRUV(dat, intra='loessShort', inter='concatenate', intra_k=5, inter_k=5); assay(dat_normalized, 'loessShort_concatenate')
Evaluation signals
- PCA plot of loessShort_concatenate assay no longer shows clustering by batch; samples cluster by biological grouping instead.
- Signal drift trend across experimental run order (batch sequence) is corrected; loess-smoothed intensity trend is flat in loessShort_concatenate vs. trending in rawImpute.
- Batch-to-batch variance (between-batch sum of squares) is reduced post-correction compared to pre-correction, quantifiable via ANOVA or batch effect variance components.
- Replicate samples (intra-batch or inter-batch) show reduced Euclidean distance in loessShort_concatenate compared to rawImpute assay.
- Metabolite abundance distributions remain stable (median, IQR unchanged); normalisation corrects systematic bias without inflating noise.
Limitations
- Requires well-designed replicate embedding; if replicates are not strategically distributed within and across batches, RUV-III decomposition may be uninformative.
- loessShort intra-batch smoothing assumes short-range (within-run) signal drift; may not correct long-range or non-monotonic instrumental artifacts.
- Concatenate inter-batch structure assumes hierarchical batch relationships; may be suboptimal if batches are unrelated or non-sequential.
- Skill is specific to metabolomics in SummarizedExperiment format; not applicable to other data types (genomics, proteomics) without adaptation.
- No changelog or version history available; reproducibility depends on hRUV package version stability (current version tested on R 4.0.3).
Evidence
- [full_text] After applying hruv normalisation with loessShort intra-batch and concatenate inter-batch parameters, the resulting loessShort_concatenate assay eliminates batch effects visible in PCA plots and corrects signal drift across experimental runs.: "After applying hruv normalisation with loessShort intra-batch and concatenate inter-batch parameters, the resulting loessShort_concatenate assay eliminates batch effects visible in PCA plots and"
- [full_text] Call the hRUV function with parameters: intra='loessShort' (intra-batch non-linear smoothing), inter='concatenate' (inter-batch hierarchical structure), intra_k=5 (short replicate neighbors for intra-batch RUV-III), inter_k=5 (batch replicate neighbors for inter-batch RUV-III).: "Call the hRUV function with parameters: intra='loessShort' (intra-batch non-linear smoothing), inter='concatenate' (inter-batch hierarchical structure), intra_k=5 (short replicate neighbors for"
- [readme] hRUV is a package for normalisation of multiple batches of metabolomics data in a hierarchical strategy with use of samples replicates in large-scale studies. The tool utilises 2 types of replicates: intra-batch and inter-batch replicates to estimate the unwanted variation within and between batches with RUV-III.: "hRUV is a package for normalisation of multiple batches of metabolomics data in a hierarchical strategy with use of samples replicates in large-scale studies. The tool utilises 2 types of replicates:"
- [intro] For intra batch normalisation, we perform loess smoothing on samples and RUV-III using short replicates with parameter k set to 5: "For intra batch normalisation, we perform loess smoothing on samples and RUV-III using short replicates with parameter k set to 5"
- [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)"
1---2name: metabolomics-batch-effect-correction3description: Use when you have log-transformed, imputed metabolomics data from multiple experimental batches (rawImpute assay) that shows visible batch effects in PCA plots or signal drift across runs, and your study design includes intra-batch replicates (technical replicates within a run) and inter-batch.4license: CC-BY-4.05---67# metabolomics-batch-effect-correction89## Summary1011Apply hierarchical RUV-III normalisation with intra-batch loess smoothing and inter-batch concatenate structure to remove batch effects and signal drift from multi-batch metabolomics data. This skill leverages sample replicates embedded within and across experimental runs to estimate and correct unwanted variation in SummarizedExperiment objects.1213## When to use1415Apply this skill when you have log-transformed, imputed metabolomics data from multiple experimental batches (rawImpute assay) that shows visible batch effects in PCA plots or signal drift across runs, and your study design includes intra-batch replicates (technical replicates within a run) and inter-batch replicates (quality control or spike-in samples across batches) that can inform RUV-III decomposition.1617## When NOT to use1819- Input data lacks intra-batch or inter-batch replicate samples; RUV-III requires both types to estimate unwanted variation.20- Metabolite filtering has not been performed; apply threshold filtering (>50% missing per batch) and intersect filtering before calling hRUV.21- Input assay is not log-transformed and imputed; hRUV expects rawImpute (log2-transformed and k-NN imputed) as input, not raw counts or unimputed data.2223## Inputs2425- SummarizedExperiment object with rawImpute assay (log-transformed and k-NN imputed)26- Batch metadata (batch assignment per sample)27- Replicate metadata (intra-batch replicate identifiers; inter-batch replicate identifiers)2829## Outputs3031- SummarizedExperiment object with new loessShort_concatenate assay (batch-corrected, normalised metabolite intensities)32- Batch effect correction applied to all metabolites quantified across all batches (intersect set)3334## How to apply3536Load a SummarizedExperiment object containing the rawImpute assay and batch/replicate metadata. Call the hRUV function with parameters: intra='loessShort' to apply non-linear smoothing within batches using RUV-III with k=5 short replicate neighbors, and inter='concatenate' to apply hierarchical inter-batch normalisation using batch replicate samples with k=5 batch replicate neighbors. Extract the resulting normalised loessShort_concatenate assay from the returned SummarizedExperiment. Evaluate success by verifying that batch structure is no longer visible in PCA plots and that signal drift across experimental runs has been corrected.3738## Related tools3940- **hRUV** (Core function implementing hierarchical RUV-III normalisation with intra-batch loess smoothing and inter-batch concatenate structure) — https://github.com/SydneyBioX/hRUV41- **RUV-III** (Statistical method underlying hRUV for estimating and removing unwanted variation using intra-batch and inter-batch replicate samples)42- **SummarizedExperiment** (Data container for storing assays, metadata, and batch/replicate annotations)43- **dplyr** (Tidyverse package for data manipulation and preparation of metadata)4445## Examples4647```48library(hRUV); dat_normalized <- hRUV(dat, intra='loessShort', inter='concatenate', intra_k=5, inter_k=5); assay(dat_normalized, 'loessShort_concatenate')49```5051## Evaluation signals5253- PCA plot of loessShort_concatenate assay no longer shows clustering by batch; samples cluster by biological grouping instead.54- Signal drift trend across experimental run order (batch sequence) is corrected; loess-smoothed intensity trend is flat in loessShort_concatenate vs. trending in rawImpute.55- Batch-to-batch variance (between-batch sum of squares) is reduced post-correction compared to pre-correction, quantifiable via ANOVA or batch effect variance components.56- Replicate samples (intra-batch or inter-batch) show reduced Euclidean distance in loessShort_concatenate compared to rawImpute assay.57- Metabolite abundance distributions remain stable (median, IQR unchanged); normalisation corrects systematic bias without inflating noise.5859## Limitations6061- Requires well-designed replicate embedding; if replicates are not strategically distributed within and across batches, RUV-III decomposition may be uninformative.62- loessShort intra-batch smoothing assumes short-range (within-run) signal drift; may not correct long-range or non-monotonic instrumental artifacts.63- Concatenate inter-batch structure assumes hierarchical batch relationships; may be suboptimal if batches are unrelated or non-sequential.64- Skill is specific to metabolomics in SummarizedExperiment format; not applicable to other data types (genomics, proteomics) without adaptation.65- No changelog or version history available; reproducibility depends on hRUV package version stability (current version tested on R 4.0.3).6667## Evidence6869- [full_text] After applying hruv normalisation with loessShort intra-batch and concatenate inter-batch parameters, the resulting loessShort_concatenate assay eliminates batch effects visible in PCA plots and corrects signal drift across experimental runs.: "After applying hruv normalisation with loessShort intra-batch and concatenate inter-batch parameters, the resulting loessShort_concatenate assay eliminates batch effects visible in PCA plots and"70- [full_text] Call the hRUV function with parameters: intra='loessShort' (intra-batch non-linear smoothing), inter='concatenate' (inter-batch hierarchical structure), intra_k=5 (short replicate neighbors for intra-batch RUV-III), inter_k=5 (batch replicate neighbors for inter-batch RUV-III).: "Call the hRUV function with parameters: intra='loessShort' (intra-batch non-linear smoothing), inter='concatenate' (inter-batch hierarchical structure), intra_k=5 (short replicate neighbors for"71- [readme] hRUV is a package for normalisation of multiple batches of metabolomics data in a hierarchical strategy with use of samples replicates in large-scale studies. The tool utilises 2 types of replicates: intra-batch and inter-batch replicates to estimate the unwanted variation within and between batches with RUV-III.: "hRUV is a package for normalisation of multiple batches of metabolomics data in a hierarchical strategy with use of samples replicates in large-scale studies. The tool utilises 2 types of replicates:"72- [intro] For intra batch normalisation, we perform loess smoothing on samples and RUV-III using short replicates with parameter k set to 5: "For intra batch normalisation, we perform loess smoothing on samples and RUV-III using short replicates with parameter k set to 5"73- [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)"
HolobiomicsLab