README.md
In CIDER: Meta-Clustering for scRNA-Seq Integration and Evaluation

CIDER

CIDER is a meta-clustering workflow designed to handle scRNA-seq data that span multiple samples or conditions. Often, these datasets are confounded by batch effects or other variables. Many existing batch-removal methods assume near-identical cell population compositions across samples. CIDER, in contrast, leverages inter-group similarity measures to guide clustering without requiring such strict assumptions.

Genome Biology (2021) publication: CIDER article
Original prototype: Hu et al., Cancer Cell 2020

Clustering: Overcome confounders in scRNA-seq data (e.g., batch effects) without requiring identical cell-type composition.
Evaluation metric: Assess whether integrated data from methods like Seurat-CCA, Harmony, or Scanorama preserve meaningful biological structure—no prior cell-type labels required.

You can install CIDER from github with:

# install.packages("devtools")
devtools::install_github('zhiyuan-hu-lab/CIDER')

If you have already integrated your scRNA-seq data (e.g., using Seurat-CCA, Harmony, or Scanorama) and want to evaluate how well the biological populations align post-integration, you can use CIDER as follows.

Before running CIDER evaluation functions, make sure that you have a Seurat object (e.g. seu.integrated) with corrected PCs in

seu.integrated@reductions$pca@cell.embeddings`

Seurat-CCA automatically put the corrected PCs there.
If other methods are used, the corrected PCs can be added using

seu.integrated@reductions$pca@cell.embeddings <- corrected.PCs

Run hdbscan clustering (optional) and compute the IDER score:

library(CIDER)
seu.integrated <- hdbscan.seurat(seu.integrated)
ider <- getIDEr(seu.integrated, verbose = FALSE)
seu.integrated <- estimateProb(seu.integrated, ider)

Visualize evaluation scores on t-SNE or UMAP:

The evaluation scores (IDER-based similarity and empirical p values) can be visualised by the scatterPlot function.

p1 <- scatterPlot(seu.integrated, "tsne", colour.by = "similarity")
p2 <- scatterPlot(seu.integrated, "tsne", colour.by = "pvalue") 
plot_grid(p1,p2, ncol = 2)

Evaluation scatterplot showing CIDER-based p-values and similarity

For a more detailed walkthrough, see the detailed tutorial of evaluation

In many scenarios, you do not start with an integrated Seurat object but still need to cluster multi-batch scRNA-seq data in a robust way. CIDER provides meta-clustering approaches:

asCIDER: When you have initial batch-specific clusters.
dnCIDER: When you do not have any initial cluster labels.

If your Seurat object (seu) has:

initial_cluster in seu@meta.data for per-batch clusters, and
Batch for batch labels,

then two main steps are:

# Step 1: Compute IDER-based similarity
ider <- getIDEr(seu, 
                group.by.var = "initial_cluster",
                batch.by.var = "Batch")

# Step 2: Perform final clustering
seu <- finalClustering(seu, ider, cutree.h = 0.45)

The final clusters will be stored in seu@meta.data$final_cluster (by default).

Detailed tutorial: asCIDER tutorial
If you do not have prior batch-specific clusters, see dnCIDER high-level or the dnCIDER detailed walk-through.

If you find CIDER helpful for your research, please cite:

Z. Hu, A. A. Ahmed, C. Yau. CIDER: an interpretable meta-clustering framework for single-cell RNA-seq data integration and evaluation. Genome Biology 22, Article number: 337 (2021); doi: https://doi.org/10.1186/s13059-021-02561-2

Z. Hu, M. Artibani, A. Alsaadi, N. Wietek, M. Morotti, T. Shi, Z. Zhong, L. Santana Gonzalez, S. El-Sahhar, M. KaramiNejadRanjbar, G. Mallett, Y. Feng, K. Masuda, Y. Zheng, K. Chong, S. Damato, S. Dhar, L. Campo, R. Garruto Campanile, V. Rai, D. Maldonado-Perez, S. Jones, V. Cerundolo, T. Sauka-Spengler, C. Yau, A. A. Ahmed. The repertoire of serous ovarian cancer non-genetic heterogeneity revealed by single-cell sequencing of normal fallopian tube epithelial cells. Cancer Cell 37 (2), p226-242.E7 (2020). doi: https://doi.org/10.1101/2021.03.29.437525