bioconductor-screclassify

---

name: bioconductor-screclassify description: A post hoc cell type classification tool to fine-tune cell type annotations generated by any cell type classification procedure with semi-supervised learning algorithm AdaSampling technique. The current version of scReClassify supports Support Vector Machine and Random Forest as a base classifier. when_to_use: Use when: An introduction to scReClassify package; Single-cell RNA-seq analysis (scReClassify). Not for: Requires R ≥ 4.1 and Bioconductor ≥ 3.16 user-invocable: false

scReClassify

Workflows

Standard Workflow

Correct and fine-tune noisy or mislabeled cell type annotations in single-cell RNA-seq data using semi-supervised learning.

library(scReClassify)
library(SingleCellExperiment)

# Load pre-processed SingleCellExperiment data
data("gse87795_subset_sce")
dat <- gse87795_subset_sce

# Perform dimension reduction using matPCs to retain PCs explaining >= 70% variance
reducedDim(dat, "matPCs") <- matPCs(dat, assay = "logNorm", 0.7)

Input: A SingleCellExperiment object containing log-normalized expression data. Output: A SingleCellExperiment object with a dimension-reduced matrix stored in reducedDim(dat, "matPCs").

When to Use

• To perform dimension reduction on single-cell RNA-seq data using matPCs to select principal components explaining at least 70% of the variance.
• To correct and fine-tune noisy or mislabeled cell type annotations in a SingleCellExperiment object.

When NOT to Use

• For initial unsupervised clustering of completely unlabeled single-cell data, use scran or Seurat because scReClassify is a post hoc semi-supervised tool that requires initial cell type annotations.
• For datasets where no initial cell type labels are available, use unsupervised clustering methods because scReClassify relies on the adaSampling algorithm which requires a set of initial (potentially noisy) labels.

Data Requirements

• Input format: SingleCellExperiment object or a matrix.
• Structure: Must contain cell type annotations in the colData (e.g., cellTypes).
• Normalization state: Log-normalized expression data (e.g., logNorm assay) is required for dimension reduction.

Key Parameters

• assay ("logNorm"): The assay name in the SingleCellExperiment object to perform dimension reduction on.

Best Practices

• Perform quality control and preprocessing of the single-cell dataset before running matPCs.
• Use matPCs to automatically select the number of principal components that explain at least 70% of the variance.

Common Pitfalls

• Running matPCs on a SingleCellExperiment without specifying the correct assay name, which will result in an error if the default assay is missing. Fix by explicitly setting the assay parameter (e.g., assay = "logNorm").

Alternatives

• Seurat for general single-cell classification and clustering.
• scater for standard PCA and dimension reduction.
• scran for unsupervised cell clustering.

Citations

• Kim T (2026), "An introduction to scReClassify package".

References

• Homepage: bioconductor.org/packages/screclassify
• Vignette: bioconductor.org/packages/release/bioc/vignettes/screclassify/inst/doc/screclassify.html