SPECK: Receptor Abundance Estimation using Reduced Rank Reconstruction and Clustered Thresholding

Description

Surface Protein abundance Estimation using CKmeans-based clustered thresholding ('SPECK') is an unsupervised learning-based method that performs receptor abundance estimation for single cell RNA-sequencing data based on reduced rank reconstruction (RRR) and a clustered thresholding mechanism. Seurat's normalization method is described in: Hao et al., (2021) doi:10.1016/j.cell.2021.04.048, Stuart et al., (2019) doi:10.1016/j.cell.2019.05.031, Butler et al., (2018) doi:10.1038/nbt.4096 and Satija et al., (2015) doi:10.1038/nbt.3192. Method for the RRR is further detailed in: Erichson et al., (2019) doi:10.18637/jss.v089.i11 and Halko et al., (2009) arXiv:0909.4061. Clustering method is outlined in: Song et al., (2020) doi:10.1093/bioinformatics/btaa613 and Wang et al., (2011) doi:10.32614/RJ-2011-015.

Author(s)

Maintainer: Azka Javaid azka.javaid.gr@dartmouth.edu

Authors:

• H. Robert Frost hildreth.r.frost@dartmouth.edu

Pipe operator

Description

See magrittr::%>% for details.

Usage

lhs %>% rhs

Arguments

Value

The result of calling 'rhs(lhs)'.

Clustered thresholding of a vector.

Description

Performs thresholding for a vector of length m from a corresponding m x n reduced rank reconstructed (RRR) matrix. Thresholding of a vector is only performed if more than one cluster is identified using the Ckmeans.1d.dp::Ckmeans.1d.dp() function based on a one-dimensional dynamic programming clustering algorithm, which functions by minimizing the sum of squares of within-cluster distances from an element to its associated cluster mean. If more than one cluster is present, then the RRR output corresponding to the nonzero elements of the least-valued cluster, as identified by the cluster mean, is set to zero. All other values in the least and higher-valued clusters are retained.

Usage

ckmeansThreshold(rrr.vector, max.num.clusters = 4, seed.ckmeans = 2)

Arguments

Value

• rrr.thresholded.vector - A thresholded vector of length m.

• num.centers - Number of identified clusters.

• max.clust.prop - Proportion of samples with the specified maximum number of clusters.

Examples

set.seed(10)
data.mat <- matrix(data = rbinom(n = 18400, size = 230, prob = 0.01), nrow = 80)
rrr.object <- randomizedRRR(counts.matrix = data.mat, rank.range.end = 60,
min.consec.diff = 0.01, rep.consec.diff = 2,
manual.rank = NULL, seed.rsvd = 1)
thresh.full.output <- ckmeansThreshold(rrr.vector = rrr.object$rrr.mat[,1],
max.num.clusters = 4, seed.ckmeans = 2)
head(thresh.full.output$rrr.thresholded.vector)
print(thresh.full.output$num.centers)
print(thresh.full.output$max.clust.prop)

Single cell RNA-sequencing (scRNA-seq) peripheral blood (PBMC) data sample.

Description

Single cell RNA-sequencing (scRNA-seq) subset of the Hao et al. 2021 human peripheral blood mononuclear cell (PBMC) data (GEO: GSE164378, DOI: 10.1016/j.cell.2021.04.048).

Usage

pbmc.rna.mat

Format

A scRNA-seq data of dgCMatrix class with 1000 rows and 33538 columns

pbmc.rna.mat

RNA expression data for 1000 cells and 33538 genes

Reduced rank reconstruction (RRR) of a matrix.

Description

Computes the rank and subsequent RRR of a m x n counts matrix. Log-normalization is first performed using the Seurat::NormalizeData() function. RRR is next performed on the normalized m x n matrix using randomized Singular Value Decomposition with the rsvd::rsvd() function. Estimated rank is selected via a construction of the standard deviations of non-centered sample principal components, which are used in a subsequent rate of change computation where each successive standard deviation value is compared to the previous to determine the rank at which the absolute value of the rate of change between consecutive values is at least 0.01 for at least two value pairs.

Usage

randomizedRRR(
  counts.matrix,
  rank.range.end = 100,
  min.consec.diff = 0.01,
  rep.consec.diff = 2,
  manual.rank = NULL,
  seed.rsvd = 1
)

Arguments

Value

• rrr.mat - A m x n RRR matrix.

• rrr.rank - Automatically computed rank.

• component.stdev - A vector corresponding to standard deviations of non-centered sample principal components.

Examples

set.seed(10)
data.mat <- matrix(data = rbinom(n = 18400, size = 230, prob = 0.2), nrow = 80)
rrr.object <- randomizedRRR(counts.matrix = data.mat, rank.range.end = 60,
min.consec.diff = 0.01, rep.consec.diff = 2,
manual.rank = NULL, seed.rsvd = 1)
print(rrr.object$component.stdev)
print(rrr.object$rrr.rank)
dim(rrr.object$rrr.mat); str(rrr.object$rrr.mat)

Abundance estimation for single cell RNA-sequencing (scRNA-seq) data.

Description

Performs normalization, reduced rank reconstruction (RRR) and thresholding for a m x n scRNA-seq matrix with m samples and n genes. The speck() function calls the randomizedRRR() function on the scRNA-seq matrix. Thresholding is next applied to each gene from the m x n RRR matrix using the ckmeansThreshold() function, resulting in a m x n thresholded matrix. See documentation for the randomizedRRR() and ckmeansThreshold() functions for individual implementation details.

Usage

speck(
  counts.matrix,
  rank.range.end = 100,
  min.consec.diff = 0.01,
  rep.consec.diff = 2,
  manual.rank = NULL,
  max.num.clusters = 4,
  seed.rsvd = 1,
  seed.ckmeans = 2
)

Arguments

Value

• thresholded.mat - A m x n thresholded RRR matrix with m samples and n genes.

• rrr.mat - A m x n RRR matrix with m samples and n genes.

• rrr.rank - Automatically computed rank.

• component.stdev - A vector corresponding to standard deviations of non-centered sample principal components.

• clust.num - A vector of length n indicating the number of clusters identified by the Ckmeans.1d.dp() algorithm for each gene.

• clust.max.prop - A vector of length n indicating the proportion of samples with the specified maximum number of clusters for each gene.

Examples

set.seed(10)
data.mat <- matrix(data = rbinom(n = 18400, size = 230, prob = 0.01), nrow = 80)
speck.full <- speck(counts.matrix = data.mat, rank.range.end = 60,
min.consec.diff = 0.01, rep.consec.diff = 2,
manual.rank = NULL, max.num.clusters = 4,
seed.rsvd = 1, seed.ckmeans = 2)
print(speck.full$component.stdev)
print(speck.full$rrr.rank)
head(speck.full$clust.num); table(speck.full$clust.num)
head(speck.full$clust.max.prop); table(speck.full$clust.max.prop)
speck.output <- speck.full$thresholded.mat
dim(speck.output); str(speck.output)