genomicInstability

Workflows

Standard Workflow

Estimate genomic instability scores and neoplastic likelihood for single cells without a prior normal reference.

library(genomicInstability)

# 1. Estimate association between gene expression and loci-blocks
cnv <- inferCNV(tpm_matrix)

# 2. Estimate the relative likelihood of cells being neoplastic
cnv <- genomicInstabilityScore(cnv)
cnv <- giLikelihood(cnv)

# 3. Plot the GIS density and fitted Gaussian models
giDensityPlot(cnv)

# 4. Refine likelihood estimation by manually specifying normal and tumor components
cnv <- giLikelihood(cnv, recompute=FALSE, normal=1, tumor=2:3)

Input: A normalized single-cell gene expression matrix (TPM); Output: An updated inferCNV object containing genomic instability scores (GIS) and relative likelihoods.

Reference Based Gia

Refine genomic instability analysis by using a subset of cells identified as normal (or an external normal dataset) as a reference null model.

library(genomicInstability)

# 1. Read User Inputs and define normal reference matrix
null_cells <- tpm_matrix[, cnv$gi_likelihood < 0.25, drop=FALSE]

# 2. Run inferCNV using the normal reference matrix
cnv_norm <- inferCNV(tpm_matrix, nullmat=null_cells)

# 3. Calculate GIS and estimate likelihood simultaneously
cnv_norm <- genomicInstabilityScore(cnv_norm, likelihood=TRUE)

# 4. Visualize the GIS density, fitted models, and null model
giDensityPlot(cnv_norm)

# 5. Plot the loci-block enrichment heatmap
plot(cnv_norm, output="heatmap.png", resolution=120, gamma=2)

Input: A normalized single-cell gene expression matrix and a designated reference matrix of normal cells; Output: A refined inferCNV object and a saved enrichment heatmap image.

When to Use

• To estimate genomic instability scores (GIS) for individual cells from single-cell RNA-seq data using genomicInstabilityScore().
• To identify putative neoplastic cells in a heterogeneous sample without requiring a prior normal reference using giLikelihood().
• To perform reference-based genomic instability analysis by specifying a normal cell matrix via the nullmat parameter in inferCNV().

When NOT to Use

• For DNA-based copy number variation detection, use packages like copynumber because genomicInstability infers genomic instability from transcriptional profiles.
• For bulk RNA-seq sample classification where single-cell resolution is not required, use standard gene set enrichment packages like GSVA because GIA is optimized for single-cell resolution.

Data Requirements

• Input Format: A standard R matrix of gene expression values (e.g., Transcripts-per-million, TPM) or a SingleCellExperiment object.
• Structure: Genes (using EntrezIDs or symbols) as rows and individual cells as columns.
• Normalization State: Fully normalized expression values (e.g., TPM) are required as input to inferCNV().

Key Parameters

• nullmat (NULL): A matrix of gene expression for normal (non-neoplastic) reference samples passed to inferCNV().
• species ("human"): Species to be analyzed (currently "human" and "mouse" are implemented) in inferCNV().
• k (100): Number of genes for each loci-block in inferCNV().
• skip (25): Displacement, in number of genes, between loci-blocks in inferCNV().
• likelihood (FALSE): Logical indicating whether to compute genomic instability likelihood simultaneously in genomicInstabilityScore().
• recompute (TRUE): Logical indicating whether to recompute the mixture model parameters in giLikelihood().
• normal (1): Integer specifying which fitted Gaussian component(s) represent normal cells in giLikelihood().
• tumor (2:3): Integer vector specifying which fitted Gaussian component(s) represent tumor cells in giLikelihood().

Best Practices

• Set a random seed using set.seed() before running downstream analyses to ensure reproducibility of the mixture model fitting.
• Inspect the raw GIS density distribution using plot(density(cnv$gis)) to verify if it exhibits a multi-modal distribution before fitting Gaussian models.
• Visualize the fitted Gaussian models using giDensityPlot() to confirm that the assigned normal and tumor components align with the observed peaks.
• Use cells with low genomic instability likelihood (e.g., gi_likelihood < 0.25) as a reference null model (nullmat) in a second iteration of inferCNV() to refine the analysis.

Common Pitfalls

• Incorrect Gaussian component assignment: Fix by manually specifying the normal and tumor component indices in giLikelihood().
• Using raw unnormalized count data: Fix by converting raw counts to TPM or another normalized format before running inferCNV().
• Mismatched gene identifiers: Fix by ensuring the input matrix uses EntrezIDs or compatible symbols that match the genomic location database for the specified species.

Alternatives

• infercnv (Bioconductor) for detailed copy number variation profiling from single-cell RNA-seq using physical genomic position maps.
• Seurat for general single-cell clustering and marker-based cell type annotation.
• scran for single-cell variance modeling and normalization.

Citations

• Puram et al. 2017, Cell (for the HNSC dataset reference).
• Alvarez et al. 2016, Nature Genetics (for the aREA algorithm reference).

References

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