scReClassify

scReClassify fine-tunes cell type annotations generated by any cell type classification procedure through semi-supervised learning to identify and reclassify potentially mislabelled cells in single-cell RNA-seq data.

Key Features:

• Semi-supervised learning framework: scReClassify employs the AdaSampling technique for post hoc cell type classification.
• Dimension reduction: The tool performs PCA as a preprocessing step prior to reclassification.
• Base classifiers: The current version supports Support Vector Machine and Random Forest as base classifiers.
• Compatibility with existing annotations: scReClassify accepts initial cell type annotations from any cell type classification procedure.

Scientific Applications:

• Cell type identification in developmental biology: scReClassify corrects mislabelled cells in scRNA-seq datasets from developmental systems.
• Cell reprogramming studies: The tool refines cell type annotations in experiments involving cellular reprogramming.
• Cancer research: scReClassify improves cell type classification accuracy in cancer scRNA-seq datasets profiling various tissues.

Methodology:

scReClassify first performs dimension reduction using PCA, then applies a semi-supervised learning method with AdaSampling to learn and reclassify cells that are likely mislabelled initially to the most probable cell types.