README.md

In EngeLab/kNNclassification: Unsupervised classification via k-Nearest-Neighbors and graph-based Louvain community detection

knnClassification

Build:

Test coverage:

Description

A package facilitating unsupervised classification via k-Nearest-Neighbors and graph-based Louvain community detection.

Installation

if(!"devtools" %in% rownames(installed.packages())) {
  install.packages("devtools")
}
devtools::install_github("EngeLab/knnClassification")

Method

The classification works by:

1. Running a principal component analysis.
2. Uses Horn’s technique to evaluate components to retain via the paran package.
3. Finds k nearest neighbors in PCA space.
4. Calculates the Euclidean distance between samples in PCA space.
5. Constructs a weighted graph where each sample is connected to the k nearest neighbors with an edge weight = 1 - Euclidean distance.
6. Uses the Louvain community detection algorithm to classify the samples.

Running the algorithm

Two datasets are included in the package which include counts and metadata for a single cell RNA sequencing experiment performed in 3 cell lines (A375, HCT116, and HOS). Using this dataset we can run the algorithm using the following commands:

library(kNNclassification)

#setup input data
s <- stringr::str_detect(colnames(testCounts), "^s")
e <- stringr::str_detect(rownames(testCounts), "^ERCC\\-[0-9]*$")
c <- testCounts[!e, s]
cpm <- t(t(c) / colSums(c) * 10^6)

#pre-run PCA
pca <- gmodels::fast.prcomp(t(cpm), scale. = TRUE)$x

#run KNN graph classification
kc <- kNNclassify(cpm, 1:nrow(c), 20, 15, pca = pca)

## 
## Using eigendecomposition of correlation matrix.
## Computing: 10%  20%  30%  40%  50%  60%  70%  80%  90%  100%

#plot
pData <- merge(kc, matrix_to_tibble(pca[, 1:2], "sample"))
plot(
  pData$PC1, pData$PC2, col = rainbow(4)[pData$louvain], pch = 16, 
  xlab = "PC1", ylab = "PC2"
)

EngeLab/kNNclassification documentation built on May 5, 2019, 9:43 p.m.