Please feel free to contact me if you have questions about these notes.

See notes grouped by date.

Benchmark principal component analysis (PCA) of scRNA-seq data in R

Principal component analysis (PCA) is frequently used for analysis of single-cell RNA-seq (scRNA-seq) data. We can use it to reduce the dimensionality of a large matrix with thousands of features (genes) to a smaller matrix with just a few factors (principal components). Since the latest scRNA-seq datasets include millions of cells, there is a need for efficient algorithms. Specifically, we need algorithms that work with sparse matrices instead of dense matrices. Here, we benchmark five implementations of singular value decomposition (SVD) and PCA.


Debug a workflow on the Terra platform

Some biomedical researchers use the Terra platform to run data analysis jobs on Google Cloud. When we run into errors, it can be daunting to figure out where the errors are coming from and how to fix them. In this tutorial, we walk through each step of viewing an error, understanding it, and fixing it in an actual workflow. I hope this illustrates the general strategy for how we can solve any workflow issue on Terra.


Tools for exploring the scientific literature

There are millions of scientific publications, and many people have created tools for exploring them. In this post, we’ll highlight a few websites and tools that allow us to search for text inside figures, view citation networks, share public comments, and more.


Find the most abundant barcodes in FASTQ files

Single-cell RNA-seq data contains oligonucleotide barcodes to uniquely identify each multiplexed sample, each single cell, and each individual molecule. Can we check which barcodes are present in a given FASTQ file? Maybe we can guess which 10x sample index was used during library preparation?


Make a table with ligands and receptors in R with OmnipathR

Curated lists of genes help computational biologists to focus analyses on a subset of genes that might be important for a research question. For example, we might be interested to focus on the genes encoding the signals and receptors for cell-to-cell communication. OmnipathR is a new R package that provides access to a vast database of genes called OmniPath, organized and curated by the Saez-Rodriguez Lab. Let’s try to use OmnipathR to create a simple table with ligands and receptors.


Resources for illustrating biomedical science

Clear illustrations are essential for effective science communication. Here, we’ll mention a few resources that might be of interest for researchers, educators, or anyone interested in making figures and presentations about biomedical science.


Make a table with your most recent coauthors in R

Some grant agencies might require a table that lists all of your coauthors, departments, and dates for publications from the last few years. Making such a table can be a laborious task for academics who publish papers with a lot of coauthors. Here, we’ll download our publication records from NCBI and use R to automatically make a table of coauthors.


Working with a sparse matrix in R

Sparse matrices are necessary for dealing with large single-cell RNA-seq datasets. They require less memory than dense matrices, and they allow some computations to be more efficient. In this note, we’ll discuss the internals of the dgCMatrix class with examples.


Harmony in motion: visualize an iterative algorithm for aligning multiple datasets

Harmony is a an algorithm for aligning multiple high-dimensional datasets, described by Ilya Korsunsky et al. in this paper. When analyzing multiple single-cell RNA-seq datasets, we often encounter the problem that each dataset is separated from the others in low dimensional space – even when we know that all of the datasets have similar cell types. To address this problem, the Harmony algorithm iteratively clusters and adjusts high dimensional datasets to integrate them on a single manifold. In this note, we will create animated visualizations to see how the algorithm works and develop an intuitive understanding.


Extract data from a PDF file with Tabula

Kirkham et al. 2006 is a prospective 2-year study of 60 patients with rheumatoid arthritis (RA). It shows that “synovial membrane cytokine mRNA expression is predictive of joint damage progression in RA”. The PDF includes a few tables with data on cytokine measurements and correlations with joint damage. Here, we’ll use Tabula to extract data from tables in the PDF file. Then we’ll make figures with R.

Make heatmaps in R with pheatmap

Here are a few tips for making heatmaps with the pheatmap R package by Raivo Kolde. We’ll use quantile color breaks, so each color represents an equal proportion of the data. We’ll also cluster the data with neatly sorted dendrograms, so it’s easy to see which samples are closely or distantly related.


Build bioinformatics pipelines with Snakemake

Snakemake is a Pythonic variant of GNU Make. Recently, I learned how to use it to build and launch bioinformatics pipelines on an LSF cluster. However, I had trouble understanding the documentation for Snakemake. I like to learn by trying simple examples, so this post will walk you through a very simple pipeline step by step. If you already know how to use Snakemake, then you might be interested to copy my Snakefiles for RNA-seq data analysis here.


Quickly aggregate your data in R with data.table

In genomics data, we often have multiple measurements for each gene. Sometimes we want to aggregate those measurements with the mean, median, or sum. The data.table R package can do this quickly with large datasets.

In this note, we compute the average of multiple measurements for each gene in a gene expression matrix.


Create a quantile-quantile plot with ggplot2

After performing many tests for statistical significance, the next step is to check if any results are more extreme than we would expect by random chance. One way to do this is by comparing the distribution of p-values from our tests to the uniform distribution with a quantile-quantile (QQ) plot. Here’s a function to create such a plot with ggplot2.


How to ssh to a remote server without typing your password

Here are a few tips to use ssh more effectively. Login to your server using public key encryption instead of typing a password. Use the ~/.ssh/config file to create short and memorable aliases for your servers. Also, use aliases to connect through a login server into a work server.