Pca

A colleague sent me a PCA plot last week. Three replicates per condition, control vs KO of a gene. PC1 cleanly separated replicate 1 from replicate 2 from replicate 3. PC2 separated control vs KO. The question that followed: should we run limma::removeBatchEffect() before differential expression? The answer is “it depends,” and most of the time the answer is “no, do not pre-correct the matrix before DE testing.” But the question hides a few subtleties worth unpacking, because I have seen people both under-correct (ignore a real block effect) and over-correct (treat true biological variation as batch and squash it).

During my work with single-cell RNA-seq data, I’ve often encountered confusion about PCA and specifically when to use the center and scale arguments in R’s prcomp() function. While tools like Seurat’s RunPCA() abstract away these details, understanding what happens under the hood is crucial for proper analysis and troubleshooting. In this post, I’ll show you exactly what center and scale do, why they matter, and what happens when you get them wrong.

To not miss a post like this, sign up for my newsletter to learn computational biology and bioinformatics. In my last blog post, I showed you how to download TCGA RNAseq count data and do PCA and make a heatmap. It is interesting to see some of the LUSC samples mix with the LUAD samples and vice versa. In this post, we will continue to use PCA to do more Exploratory data analysis (EDA).

To not miss a post like this, sign up for my newsletter to learn computational biology and bioinformatics. In my last post, I showed you how to use PCA for bulk RNAseq data. Today, let’s see how we can use it for scATACseq data. Download the example dataset from 10x genomics https://support.10xgenomics.com/single-cell-atac/datasets/1.1.0/atac_pbmc_5k_v1 The dataset is 5k Peripheral blood mononuclear cells (PBMCs) from a healthy donor (v1.0). Download the atac_pbmc_5k_v1_filtered_peak_bc_matrix.tar.gz file and unzip it.

To not miss a post like this, sign up for my newsletter to learn computational biology and bioinformatics. what is PCA? Principal Component Analysis (PCA) is a mathematical technique used to reduce the dimensionality of large datasets while preserving the most important patterns in the data. It transforms the original high-dimensional data into a smaller set of new variables called principal components (PCs), which capture the most variation in the data.

To not miss a post like this, sign up for my newsletter to learn computational biology and bioinformatics. Understand CCA Following my last blog post on PCA projection and cell label transfer, we are going to talk about CCA. In single-cell RNA-seq data integration using Canonical Correlation Analysis (CCA), we typically align two matrices representing different datasets, where both datasets have the same set of genes but different numbers of cells.

To not miss a post like this, sign up for my newsletter to learn computational biology and bioinformatics. Understand the example datasets We will use PBMC3k and PBMC10k data. We will project the PBMC3k data to the PBMC10k data and get the labels library(Seurat) library(Matrix) library(irlba) # For PCA library(RcppAnnoy) # For fast nearest neighbor search library(dplyr) # Assuming the PBMC datasets (3k and 10k) are already normalized # and represented as sparse matrices # devtools::install_github('satijalab/seurat-data') library(SeuratData) #AvailableData() #InstallData("pbmc3k") pbmc3k<-UpdateSeuratObject(pbmc3k) pbmc3k@meta.

PCA is a critical method for dimension reduction for high-dimensional data. High-dimensional data are data with features (p) a lot more than observations (n). However, this is changing with single-cell RNAseq data. Now, we can sequence millions (n) of single cells and each cell has ~20,000 genes/features (p). I suggest you read my previous blog post on using svd to calculate PCs. Single-cell expression data PCA In single-cell RNAseq analysis, feature selection will be performed first.

PCA in practice. Principal Component Analysis(PCA) is a very important skill for dimention reduction to analyze high-dimentional data. High-dimentional data are data with features (p) a lot more than observations (n). This types of data are very commonly generated from high-throuput sequencing experiments. For example, an RNA-seq or microarry experiment measures expression of tens of thousands of genes for only 8 samples (4 controls and 4 treatments). Let’s use a microarray data for demonstration.

In my last blog post, I showed an example to use heatmap to represent discrete values. I am going to continue the theme to introduce heatmap to represent…