bioconductor-awst

name

bioconductor-awst

description

We propose an Asymmetric Within-Sample Transformation (AWST) to regularize RNA-seq read counts and reduce the effect of noise on the classification of samples. AWST comprises two main steps: standardization and smoothing. These steps transform gene expression data to reduce the noise of the lowly expressed features, which suffer from background effects and low signal-to-noise ratio, and the influence of the highly expressed features, which may be the result of amplification bias and other experi

awst

Workflows

Standard Workflow

library(awst)
library(SummarizedExperiment)
library(EDASeq)

# Load data (e.g., SummarizedExperiment)
data(airway, package = "airway")

# Filter lowly expressed genes
filter <- rowMeans(assay(airway)) >= 10
se <- airway[filter,]

# Apply Asymmetric Within-Sample Transformation
se <- awst(se)

# Filter uninformative genes using entropy
filtered <- gene_filter(se)

# Optional: Apply AWST after full-quantile normalization
assay(se, "fq") <- betweenLaneNormalization(assay(se), which = "full")
se_normalized <- awst(se, expr_values = "fq")

Input/Output Note: Inputs a SummarizedExperiment containing raw RNA-seq counts; outputs a SummarizedExperiment with transformed values in the "awst" assay and optionally filtered features.

When to Use

• To regularize RNA-seq read counts and reduce the effect of noise on sample clustering or classification, especially for highly degraded or low input material samples like tumors or single cells.
• To perform per-sample standardization and asymmetric winsorization using awst() to reduce the influence of highly expressed features and noise of lowly expressed features.
• To filter out uninformative genes that contribute only noise to sample distances using the entropy-based gene_filter() function.

When NOT to Use

• For differential expression analysis where raw or specifically modeled counts are required (use DESeq2 or edgeR instead).
• For multi-sample batch effect correction across highly heterogeneous batches without prior normalization (use sva::ComBat or removeBatchEffect from limma instead).

Data Requirements

• Input must be a SummarizedExperiment object (or a matrix of counts).
• Contains raw or normalized RNA-seq read counts (e.g., from airway dataset).
• Lowly expressed genes should be pre-filtered (e.g., keeping genes with rowMeans(assay(se)) >= 10).

Key Parameters

• expr_values ("counts"): Character string specifying which assay of the input SummarizedExperiment to use for the transformation (e.g., "fq" after running betweenLaneNormalization).

Best Practices

• Pre-filter non-expressed or lowly expressed genes (e.g., average count < 10) before applying AWST to reduce computational load and noise.
• Apply a prior normalization step, such as full-quantile normalization via betweenLaneNormalization(), which allows awst() to estimate parameters only once for all samples.
• Use gene_filter() after awst() to remove uninformative genes and improve sample separation in downstream analyses like PCA (prcomp()).

Common Pitfalls

• Pitfall: Running awst() on log-transformed data. Fix: Ensure the input assay contains non-log-transformed counts, as awst() internally exploits the log-normal probability distribution.
• Pitfall: Retaining too many uninformative genes which dilutes the biological signal in PCA. Fix: Run gene_filter() on the transformed object to retain only genes contributing to sample distance.

Alternatives

• DESeq2 for variance stabilizing transformation (VST) or rlog.
• edgeR for calculating CPM/RPKM or TMM normalization.
• scater / scran for single-cell specific normalization and pooling.

Citations

• Risso D, Pagnotta SM (2021). "Per-sample standardization and asymmetric winsorization lead to accurate clustering of RNA-seq expression profiles." Bioinformatics. doi:10.1093/bioinformatics/btab091.

References

• Homepage: https://bioconductor.org/packages/awst
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/awst/inst/doc/awst_intro.html