In CompEpigen/scMethrix: Fast and efficient summarization of single cell whole genome bisulfate data

Clustering data

Unlike bulk sequencing, where the measurements are averaged over many cells, single-cell sequencing allows the identification of cell type for individual cells. This is a common step in many single-cell expression workflows. Multiple algorithms and distance metrics are included in the package for maximum flexibility.

Distance metrics

Most algorithms operate via a distance metric, an all-by-all matrix where each sample is numerically compared to each other sample. Samples with few methylated CpG sites in common will have a large distance metric, where as two samples with identical methylation will have a distance metric of zero.

Available distance metrics include pearson, spearman, and tau (all via bioDist); and euclidean, maximum, manhattan, canberra, binary, and minkowski (all via stats::dist).

scMethrix::get_distance_matrix(scm, assay = "score" , type = "pearson")

Additional distance metrics can be used via an arbitrary function input. The input must the assay matrix and output must be an all-by-all matrix filled with distance values for each respective pair. It will be internally cast with as.dist() before computation.

fun <- function(mtx) dist(mtx, method = "euclidean")
scMethrix::get_distance_matrix(scm, assay = "score" , type = fun)

Clustering

Clustering algorithms include hierarchical (via stats::hclust), partition (via stats::kmeans), and model-based (via mclust::Mclust). Distance metrics are not used for model-based clustering. The identified clusters will be stored in the colData slot of the output scMethrix object.

scm.cluster <- scMethrix::cluster_scMethrix(scm, assay = "score", dist = dist, n_clusters = 2, colname = "Heirarchical", type="hierarchical")
colData(scm.cluster)

Like the distance metric, an arbitrary clustering algorithm can be used. It must accept a dist object, and return a data.frame with two columns named "Sample" and "Cluster". The column "Cluster" will be renamed by the value in colname before returning, if given.

fun <- function (dist) {
  fit <- stats::hclust(dist, method="ward.D")
  fit <- stats::cutree(fit, k=ncol(dist)/2)
  colData <- data.frame(Sample = names(fit), Cluster = fit)
  colData
}

scm.cluster <- scMethrix::cluster_scMethrix(scm, assay = "score", dist = dist, n_clusters = 2, colname = "Heir", type=fun)
colData(scm.cluster)

Visualization

For visualizing the clustering, the generated colData can be used for annotation in dimensionality reduction plots.

scm <- scMethrix::dim_red_scMethrix(scm,type="PCA",top_var = 10000)
plot_dim_red(scm,type="PCA",shape_anno = "Cluster",color_anno="Cluster")

CompEpigen/scMethrix documentation built on Nov. 6, 2021, 3:09 p.m.