bioconductor-sgcp

name

bioconductor-sgcp

description

SGC is a semi-supervised pipeline for gene clustering in gene co-expression networks. SGC consists of multiple novel steps that enable the computation of highly enriched modules in an unsupervised manner. But unlike all existing frameworks, it further incorporates a novel step that leverages Gene Ontology information in a semi-supervised clustering method that further improves the quality of the computed modules.

SGCP

Workflows

Standard Workflow

Construct gene co-expression networks, perform clustering, and apply semi-supervised classification using Gene Ontology information.

library(SGCP)
library(SummarizedExperiment)
library(org.Hs.eg.db)

# Load example dataset
data(cheng)
expData <- assay(cheng)
geneID <- rowData(cheng)$ENTREZID
annotation_db <- "org.Hs.eg.db"

# Option 1: Automatic Run using ezSGCP
# (Precomputed data loaded here to avoid long execution times)
data(sgcp)
summary(sgcp, show.data=TRUE)

# Plot PCA and silhouette index
SGCP_ezPLOT(sgcp = sgcp, expreData = expData, silhouette_index = TRUE, keep = FALSE)

# Option 2: Step-by-Step Run
# 1. Network Construction
resAdja <- adjacencyMatrix(expData = expData, hm = NULL)

# 2. Network Clustering (using precomputed resClus for speed)
data(resClus)
summary(resClus)

# Plot PCA of expression data
pl <- SGCP_plot_pca(m = expData, clusLabs = NULL, tit = "PCA plot", ps = .5)
print(pl)

Note on inputs/outputs:

• Input: A gene expression matrix, a vector of gene Entrez IDs, and an organism-specific annotation database.
• Output: A list containing initial clusters, semi-labeled genes, semi-supervised classification results, and final biologically enriched gene modules.

When to Use

• Gene Co-expression Networks: Constructing gene co-expression networks and organizing genes into modules using adjacencyMatrix() and clustering().
• GO Integration: Integrating Gene Ontology (GO) enrichment analysis into the clustering pipeline using geneOntology().
• Semi-supervised Classification: Applying semi-supervised classification (e.g., KNN or logistic regression) to refine gene modules using semiSupervised().
• Network Visualization: Visualizing gene expression PCA plots and network adjacency heatmaps using SGCP_plot_pca() and SGCP_plot_heatMap().

When NOT to Use

• Raw Sequence Alignment: For raw sequence alignment or differential expression analysis, use packages like DESeq2 or edgeR because SGCP requires preprocessed, normalized expression data with non-zero variance.
• Unsupervised Clustering Without GO: For purely unsupervised clustering without any biological annotation or GO integration, use standard R clustering functions (e.g., kmeans or hclust) because SGCP is specifically designed to leverage GO information.

Data Requirements

• Expression Data (expData): A matrix or data frame of size $m \times n$ (genes by samples) of normalized DNA-microarray or RNA-seq data. Genes must have expression values across all samples (no missing values) and non-zero variance.
• Gene Identifiers (geneID): A vector of size $m$ containing unique gene identifiers (e.g., Entrez IDs) compatible with the annotation database.
• Annotation Database (annotation_db): A genome-wide annotation package name (e.g., "org.Hs.eg.db").

Key Parameters

• calibration (FALSE): Boolean in adjacencyMatrix() to perform network calibration.
• norm (TRUE): Boolean in adjacencyMatrix() to divide each gene vector by its L2 norm.
• tom (TRUE): Boolean in adjacencyMatrix() to add Topological Overlap Matrix to the network.
• kopt (NULL): Integer in clustering() specifying the user-defined optimal number of clusters.
• method (NULL): Method for determining the number of clusters in clustering(); options include "relativeGap", "secondOrderGap", and "additiveGap".
• direction (c("over", "under")): Test direction for GO enrichment in geneOntology().
• ontology (c("BP", "CC", "MF")): GO ontologies to test in geneOntology().
• model ("knn"): Classification model in semiSupervised(); options include "knn" and "lr".

Best Practices

• Ensure all preprocessing, normalization, and batch effect corrections are completed on the expression matrix before inputting to SGCP.
• Filter out genes with zero variance or missing values across samples prior to running adjacencyMatrix().
• Evaluate and compare the three cluster-number determination methods ("relativeGap", "secondOrderGap", "additiveGap") using GO validation.
• Enable the silhouette index calculation (sil = TRUE) in clustering() to evaluate the quality of the initial clusters.

Common Pitfalls

• Mismatched Gene Identifiers: Mismatch between gene identifiers in geneID and the standard expected by GOstats / annotation_db. Fix: Ensure geneID contains valid Entrez IDs (or TAIR IDs for Arabidopsis) that match the selected annotation_db.
• Long Computation Times: Long computation times during the GO validation or clustering steps. Fix: Use precomputed results where possible, or reduce the number of genes to those with the highest variance.
• Zero Variance Genes: Error due to genes with zero variance across samples. Fix: Filter the input expression matrix to remove zero-variance genes before running the pipeline.

Alternatives

• WGCNA for standard unsupervised gene co-expression network analysis.
• GOstats for standalone hypergeometric testing of Gene Ontology terms.
• SummarizedExperiment for general containerization of biological assays and genomic features.

Citations

• Cheng, J. et al. (Ischemic cardiomyopathy dataset reference).

References

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