Dino

Workflows

Standard Workflow

Filter and normalize a raw UMI count matrix using Dino, then convert it to a Seurat object for downstream clustering and visualization.

library(Dino)
library(Seurat)
library(Matrix)

# Load dataset
data("pbmcSmall")

# 1. Filter out genes with low non-zero expression
pbmcSmall <- pbmcSmall[rowSums(pbmcSmall != 0) >= 20, ]

# Normalize data
pbmcSmall_Norm <- Dino(pbmcSmall)

# 2. Run downstream Seurat clustering and visualization
pbmcSmall_Seurat <- SeuratFromDino(pbmcSmall_Norm, doNorm = FALSE)
pbmcSmall_Seurat <- FindVariableFeatures(pbmcSmall_Seurat, selection.method = "mvp")
pbmcSmall_Seurat <- ScaleData(pbmcSmall_Seurat, features = rownames(pbmcSmall_Norm))
pbmcSmall_Seurat <- RunPCA(pbmcSmall_Seurat, features = VariableFeatures(object = pbmcSmall_Seurat), verbose = FALSE)
pbmcSmall_Seurat <- FindNeighbors(pbmcSmall_Seurat, dims = 1:10)
pbmcSmall_Seurat <- FindClusters(pbmcSmall_Seurat, verbose = FALSE)
pbmcSmall_Seurat <- RunUMAP(pbmcSmall_Seurat, dims = 1:10)
DimPlot(pbmcSmall_Seurat, reduction = "umap")

Note on inputs/outputs: Input is a raw UMI count matrix (pbmcSmall), and output is a fully processed and clustered Seurat object.

Single Cell Experiment Normalization

Normalize scRNA-seq data formatted as a SingleCellExperiment object using Dino.

library(SingleCellExperiment)
library(Dino)

# Load dataset
data("pbmcSmall")

# Create a SingleCellExperiment object from the count matrix
pbmc_SCE <- SingleCellExperiment(assays = list("counts" = pbmcSmall))

# Run Dino
pbmc_SCE <- Dino_SCE(pbmc_SCE)

# Inspect or extract the normalized counts using normcounts
normalized_counts <- normcounts(pbmc_SCE)
str(normalized_counts)

Note on inputs/outputs: Input is a SingleCellExperiment object containing raw counts, and output is the same object with normalized counts accessible via normcounts().

Custom Depth Normalization

Normalize a UMI count matrix using Dino with custom sequencing depths calculated via Scran size factors.

library(Dino)
library(scran)

# Load dataset
data("pbmcSmall")

# Compute size factors using Scran's calculateSumFactors
scranSizes <- calculateSumFactors(pbmcSmall)

# Re-normalize data with custom depths
pbmcSmall_SNorm <- Dino(pbmcSmall, nCores = 1, depth = log(scranSizes))

Note on inputs/outputs: Input is a raw UMI count matrix, and output is a normalized sparse matrix using custom log-transformed size factors.

When to Use

• Normalizing high-throughput single-cell RNA-sequencing (scRNA-seq) UMI count data to correct for technical variation in sequencing depth using Dino().
• Integrating normalized expression matrices directly into Seurat workflows using SeuratFromDino().
• Normalizing data structured as a SingleCellExperiment object using Dino_SCE().

When NOT to Use

• For non-UMI sequencing protocols (e.g., traditional bulk RNA-seq), use packages like DESeq2 or edgeR because Dino is specifically designed and optimized for UMI count data.
• For correcting batch effects or other sample-specific effects, use packages like Combat or harmony because Dino only corrects for technical variation correlated with sequencing depth.

Data Requirements

• Input format: Sparse or dense matrix of expression (e.g., dgCMatrix or standard matrix) with genes (features) on the rows and cells (samples) on the columns.
• Structure: Non-negative integer UMI counts (e.g., pbmcSmall).
• Normalization state: Raw, un-normalized count data.

Key Parameters

• nCores (2): Number of parallel instances/threads to run. Setting to 0 utilizes all available cores.
• depth (NULL): Alternate cell-specific sequencing depths (e.g., log of scran size factors).
• doNorm (TRUE): Logical indicating whether to perform normalization inside SeuratFromDino().
• doLog (TRUE): Logical indicating whether to log-transform expression as log(x + 1) in SeuratFromDino().

Best Practices

• Filter out genes lacking a minimum of non-zero expression (e.g., keeping genes with at least 20 non-zero samples using rowSums(pbmcSmall != 0) >= 20) prior to normalization to improve downstream analysis.
• Set a random seed using set.seed() before running Dino to ensure reproducibility of the distributional resampling.
• Ensure that when providing custom sequencing depths via depth, the median depth is close to 1.

Common Pitfalls

• Running Dino on non-UMI datasets: This can lead to poor normalization because Dino's Negative Binomial mixture model is specifically tailored for sparse UMI data. Fix by using Dino only on UMI counts.
• Failing to filter low-expression genes: Dino will default to simple scaling with sequencing depth for genes with fewer than 10 non-zero samples. Fix by filtering genes with low non-zero expression (e.g., rowSums(counts != 0) >= 20) beforehand.

Alternatives

• Seurat: For comprehensive single-cell workflows including alternative normalization methods like SCTransform.
• scran: For pooling-based normalization of single-cell data using calculateSumFactors.
• scater: For quality control and visualization of single-cell RNA-seq data.

Citations

• Brown, J., Ni, Z., Mohanty, C., Bacher, R., and Kendziorski, C. (2021). "Normalization by distributional resampling of high throughput single-cell RNA-sequencing data." Bioinformatics, 37, 4123-4128.
• Lun, A. T. L., Bach, K. and Marioni, J. C. (2016). "Pooling across cells to normalize single-cell RNA sequencing data with many zero counts." Genome Biol., 17, 1–14.

References

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