---
title: "scISR package manual"
author: "Duc Tran"
date: "`r Sys.Date()`"
output:
  html_document:
    df_print: paged

editor_options:
  chunk_output_type: inline
vignette: >
  %\VignetteIndexEntry{scISR package manual}
  %\VignetteEngine{knitr::knitr}
  %\usepackage[UTF-8]{inputenc}
---


# Introduction
Recent advances in biochemistry and single-cell RNA sequencing (scRNA-seq) have allowed us to monitor the biological systems at the single-cell resolution. However, the low capture of mRNA material within individual cells often leads to inaccurate quantification of genetic material. Consequently, a significant amount of expression values are reported as missing, which are often referred to as dropouts. These missing information can heavily impact the accuracy of downstream analyses.

To addreess this problem, we introduce a novel inputation method, names single-cell Imputation via Subspace Regression (scISR). scISR performs imputation for single-cell sequencing data. scISR identifies the true dropout values in the scRNA-seq dataset using hyper-geomtric testing approach. Based on the result obtained from hyper-geometric testing, the original dataset is segregated into two including training data and imputable data. Next, training data is used for constructing a generalize linear regression model that is used for imputation on the imputable data.  

# Installation
The `scISR` package can be installed from GitHub using `devtools` package
```{r, eval=FALSE}
# Install devtools package
utils::install.packages('devtools')
# Install scISR from GitHub
devtools::install_github('duct317/scISR')
```

# Using scISR
## Preparing data
Load the example, Goolam dataset. The data is a list consists of two elements: data matrix with rows as genes and columns as samples, and a vector with the cell types information.
```{r, eval=FALSE}
#Load required library
library(scISR)

# Load example data (Goolam dataset with reduced number of genes), other dataset can be download from our server at http://scisr.tinnguyen-lab.com/
data('Goolam')
# Raw data
raw <- Goolam$data
# Cell types information
label <- Goolam$label
```

## Imputation
We can use the main funtion `scISR` to impute the raw data. 
```{r, eval=FALSE}
# Generating subtyping result
set.seed(1)
imputed <- scISR(data = raw, ncores = 4)
```

## Result evaluation
The quality of the imputation can be evaluated using clustering analysis. We can evaluate the accuracy of clusters obtained from clustering analysis using raw and imputed data with adjusted Rand Index (ARI). Higher ARI means higher agreement between a given clustering and the ground truth. The clusters from imputed data have much higher accuracy. 
```{r, eval=FALSE}
library(irlba)
library(mclust)
# Perform PCA and k-means clustering on raw data
set.seed(1)
# Filter genes that have only zeros from raw data
raw_filer <- raw[rowSums(raw != 0) > 0, ]
pca_raw <- irlba::prcomp_irlba(t(raw_filer), n = 50)$x
cluster_raw <- kmeans(pca_raw, length(unique(label)),
                      nstart = 2000, iter.max = 2000)$cluster
print(paste('ARI of clusters using raw data:', round(adjustedRandIndex(cluster_raw, label),3)))

# Perform PCA and k-means clustering on imputed data
set.seed(1)
pca_imputed <- irlba::prcomp_irlba(t(imputed), n = 50)$x
cluster_imputed <- kmeans(pca_imputed, length(unique(label)),
                          nstart = 2000, iter.max = 2000)$cluster
print(paste('ARI of clusters using imputed data:', round(adjustedRandIndex(cluster_imputed, label),3)))
```

```{r, eval=FALSE}
sessionInfo()
```