Data Science to Analyze Big Genomic Data

July 28, 2018

(This article was first published on R – NYC Data Science Academy Blog, and kindly contributed to R-bloggers)

Finding the neural stem cell populations in mouse brain



The main objective of this project is to identify new stem cell populations in mouse brain. Characterizing the unique gene expression signature of these cells could be the starting point for finding and defining cancer stem cells (CSC) in human tumors. In particular, glioblastoma, which is the most common malignant brain tumor diagnosed in adults. Despite numerous advances in cancer therapy, these tumors remain incurable, suggesting that current treatments fail to target the cells responsible for tumor growth. One possible explanation for this would be that CSCs are resistant to current therapies and responsible for tumor remission.

In order to develop new therapies that target CSC, reliable marker genes (a marker gene is a gen that is uniquely expressed in a given cell type) that define the “CSC signature” (a molecular signature is a group of marker genes of a given cell type), are needed. The first step is to study mouse brain and define their Neural Stem Cell (NSC) signature. Based on the mouse NSC Signature we would define the Cancer stem signature in human tumors. And finally we would use the Cancer Stem Cell marker genes to screen for new drugs that could lead to novel Glioblastoma treatments.

The dataset that I used for this project is the result of dissecting a specific part of mice brain (dentate gyrus, responsible for learning and memory), dissociating the cells into a single cell suspension and performing Next Generation Sequencing on individual cells. In particular, single cell RNA sequencing. This technique allows the quantification of all the RNA transcript of each individual cell. We quantify RNA instead of DNA because the DNA content is the same in every cell of our body, however the RNA expression profile uniquely defines a cell’s identity and function, it is what distinguish a skin cell from a neuron.

After filtering low quality cells, the final data object had 18572 rows and 17922 columns. Each row corresponds to a gene and each column corresponds to a cell. The values in the database are a measure of the RNA expression of each gene in each cell. Each cell expresses an average of 2048 genes. The size of next generation sequencing genomic data, in this case the processed database consumed 3.2Gb of memory, makes Data Science and machine learning algorithms indispensable tools for handling big data.


Who is who

After filtering the data, I performed a clustering analysis based on K-nearest neighbor KNN algorithm. The resulting plots are called t-SNE (T-distributed Stochastic Neighbor Embedding) plots. In these plots each cell is represented by a dot (hence each dot summarizes the information of all the rows and columns referred to the expression levels for one single cell). The spatial proximity between the dots is related to the molecular similarities. Therefore, cells that cluster together belong to the same cell type.

In order to define “who is who” and where is our stem cell populations we have to interrogate the gene expression of the different clusters. For example the genes Gfap and Id4 are known to be expressed in astrocytes and stem cells, therefore we know that the highlighted groups are astrocytes and stem cells. Other genes such as Rbfox3 and Syt1 are known to be expressed in neurons therefore we know the highlighted clusters belong to neurons.

Going through different gene expression profiles, with the help of this interactive tSNE plot and dot plot, we figure out which cluster belongs to which cell type and most importantly, where our stem cell clusters are. In particular, I was able to find a new SC cluster, here called RGL-1 that is related to the previously identified RGL-2.


Developmental time

After clustering, I used machine learning algorithms to further determine the relative location of each cluster in the neurogenic differentiation trajectory. I used semi supervised ordering mode to order the cells along the trajectory based on known marker genes that define the biological progress through the neuronal differentiation process and other (1000) genes that covary with them. Once the algorithm has learned the specific trajectory of expression changes, it places each cell at its proper position in the trajectory all the way from the most primitive stem cell state to a more differentiated neuroblast, going through the intermediate NPC (Neural Progenitor Cell) state.

The left panel shows the stem cell to neuron differentiation trajectory and right panel shows the expression of specific genes along developmental Pseudotime. As seen in the plot below, stem cell genes like Gfap and Id4 start high and end low.  When looking for neurogenic genes such as Dcx and Calb2 we can verify that they start low and end up high as expected.

In addition to finding the “developmental” order of our clusters, I could see that the newly discovered stem cell cluster (RGL-1) starts at the beginning of the trajectory, supporting that it indeed corresponds to a stem cell cell population.



So far, we have identified the different cell types, including the stem cell populations, we have reconstructed developmental time and finally it is time to find new genes, new NSC markers.

In order to find new genes expressed in our clusters, we use heatmaps. The red color indicates highly expressed genes and the blue color indicates low expressed genes. The graph shows the top 15 genes (highly expressed) on the right hand side of the graph per cluster. In the heatmap we look for genes that are highly expressed in the stem cell populations (RGL-1 and RGL-2) but not in the other cell types. Genes that are uniquely expressed in the stem cell clusters are ideal candidates for new stem cell marker genes. Ultimately, NSC marker genes could be the first step towards the discovery and validation of CSC marker genes in human tumors. Lastly, CSC marker genes would be used to screen for new drugs that could lead to the development of new therapies for glioblastoma.


  • All the cell types in this particular region of mouse brain were identified based on expression of known marker genes
  • A new neural stem cell population (RGL-1) was identified and characterized
  • The newly defined stem cell group is located at the beginning of the differentiation trajectory
  • Uniquely expressed genes were identified in a new stem cell population that are ideal candidates for new stem cell markers

Disclaimers and Acknowledgements

This blog includes unpublished data and is not intended to be used for submissions or citations.

The data presented in this blog was generated in Luis Parada’s laboratory at Memorial Sloan Kettering Cancer Center. Single cell RNA sequencing (10X Genomics Chromium) was performed at the Epigenomics Core at Weill Cornell Medicine.

Seurat package for R was used for the pre-processing and clustering and Monocle package for R was used for pseudotime analysis.

Featured image is a courtesy of National Human Genome Research Institute (NHGRI).

To leave a comment for the author, please follow the link and comment on their blog: R – NYC Data Science Academy Blog. offers daily e-mail updates about R news and tutorials on topics such as: Data science, Big Data, R jobs, visualization (ggplot2, Boxplots, maps, animation), programming (RStudio, Sweave, LaTeX, SQL, Eclipse, git, hadoop, Web Scraping) statistics (regression, PCA, time series, trading) and more...

If you got this far, why not subscribe for updates from the site? Choose your flavor: e-mail, twitter, RSS, or facebook...

Comments are closed.

Search R-bloggers


Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)