Skip to main content

BF528 - Applications in Translational Bioinformatics

Final Project - single cell RNAseq

For the final project, you may use either of the two provided datasets or choose one of your own interest provided you get approval.

Paper 1: https://www.nature.com/articles/s42003-021-02562-8

Paper 2: https://www.nature.com/articles/s41467-023-40727-7

Generally speaking, the minimum requirements for the datasets are the following:

  1. The dataset must consist of single cell or single nucleus sequencing data
  2. It must consist of at least 4 replicates / samples

You are responsible for obtaining the data in a usable format for either of the provided papers or your own. Remember that the format of the submitted data may vary depending on the lab itself, the journal it was published in, or even the requirements of the grant mechanism that funded the study. For some datasets, it may be as simple as finding the GEO accession and downloading the alignment results directly or you may need to download the BAM / FASTQ files are process them.

By now, we will have gone through the basic workflow for a single cell dataset in-class. For the final project, you will be processing single cell RNAseq data from a published experiment on your own and focusing on some of the more common biological analyses performed on scRNAseq.

These directions will give you a basic scaffold of what you need to do and the rest you will need to implement on your own. You will be asked to produce certain figures / plots and write a short discussion on what is shown, your interpretation of the findings and how it affects how you will proceed with your analysis. Please create a single jupyter notebook or Rmarkdown that contains both the required plots and answers to the discussion questions.

If you use a jupyter notebook (scanpy), please also include a .YML file with the list of packages and versions you used. If you use a Rmarkdown (seurat), please include a printout from the sessionInfo() function in R with all of your utilized packages.

Please either knit your Rmarkdown to a HTML or render your jupyter notebook to a HTML report

Preprocessing

Perform appropriate QC for each of the samples in your dataset separately. In addition to the basic QC metrics, you should utilize a doublet detection tool of your choice.

Figures

  1. Create a plot or plots that contain at minimum information on the number of unique genes detected per barcode, the total number of molecules per barcode, and the percentage of reads that map to the mitochondrial genome.

  2. An integrated plot that visualizes the above metrics jointly for each sample.

  3. Results from the doublet detection algorithm you employed

  4. Create a table that displays how many cells / genes are present in your dataset before and after your filtering thresholds are implemented

Discussion

In your discussion, make sure to address the following points / questions:

  • What filtering thresholds did you choose and how did you decide on them?
  • How many cells / genes are present before and after implementing your filtering thresholds?
  • Look in the literature, what are some potential strategies to set thresholds that don’t rely on visual inspection of plots?

Count Normalization

Discussion

Choose a method to normalize the scRNAseq data and ensure that you explain the exact normalization procedure.

Feature Selection

Choose an appropriate method for feature selection and the number of highly variable features you will use in downstream analyses.

Figures

  • Create a plot of the highly variable features by your chosen metric.

Discussion

  • Explain the method used to determine highly variable features and report how many variable features you chose to use for downstream dimensional reduction.

  • Include a brief statement on how many genes meet your chosen threshold to be considered highly variable and how many are not considered

PCA

Perform dimensional reduction using PCA on your highly variable features.

Figures

  • Create a plot that justifies your choice of how many PCs to utilize in downstream analyses

Discussion

  • Justify your choice in writing and briefly remark on your plot.

Clustering and visualization

Decide on an appropriate method to cluster your cells and create a visualization of the 2D embedding (UMAP / T-SNE). Choose an appropriate resolution for the clustering algorithm.

Figures

  • Create a plot visualizing the clustering results of your cells. Ensure that it has labels that identify the different clusters as determined by the unsupervised clustering method you employed

  • Create a plot visualizing the clustering results of your cells. Ensure that it has labels that identify which samples each cell originated from.

Discussion

  • Write a brief paragraph describing the results up to this point. In it, ensure that you include the following information:

    1. How many cells come from each sample individually? How many total cells present in the entire dataset?
    2. How many clusters are present?
    3. What clustering resolution did you use?

  • Use the second plot you created and briefly remark on whether you will perform integration.

Integration (optional)

If you determined from your clustering that the dataset would benefit from integration, please choose an appropriate method and apply it to your dataset.

Figures

  • Create a plot visualizing the clustering results of your cells before and after integration. Ensure that it has labels that identify which samples each cell originated from.

Discussion

  • Briefly remark on the plots and the effects you observe due to the integration

Marker Gene Analysis

Perform marker gene analysis using a method of your choice.

Figures

  • Create a table listing the top five marker genes for each of your clusters

Discussion

  • Briefly describe the method performed to identify marker genes. Discuss a few advantages and disadvantages of the method used to perform marker gene analysis.

Automatic Annotation of Cell labels

Use an annotation algorithm to assign preliminary labels for the cell identities of your clusters. You may use your choice of tool as long as it has a documented code base hosted publicly and has an associated publication describing its use.

Figures

  • Create a visualization of your cell clustering with the labels assigned by the results of running the labeling tool of your choice.

Discussion

  • Briefly remark on how the algorithm works and ensure you cite the original publication. You may paraphrase information from the publication.

  • Comment on the cell identities of your clusters and any speculation as to the original source tissue of the samples based on what you see

Manual Cluster Labeling

Perform literature searches using the marker gene analysis you performed before. In combination with the results from your cell annotation algorithm, provide a final determination of the cell identities for each of your clusters.

Figures

  • Create a single plot / figure that displays the top marker genes for every cluster in your dataset. This plot may be in any style (a heatmap, a tracksplot, violin plot, etc.)

  • For the top three clusters with the most cells, create individual plots of at least 5 of the defining marker genes for that cluster and their expression compared across all clusters

  • Create a visualization / plot of your clusters with your manual annotations and labels of cell identities.

Discussion

  • Write a paragraph describing your cell annotations and explaining what cell label you have assigned to each cluster. Ensure that you have at least one citation that supports your labeling and choice of marker genes for each cluster.

0.5 Replicate two of the major findings / analyses from the publication

Please look in the publication for the dataset and choose two of their major findings / experiments and attempt to replicate them with your own workflow.

These must be different and can involve any of the following types of analyses:

  1. Pseudotime analysis (Monocle, Slingshot, CellRank, etc.)
  2. RNA velocity analysis (scVelo, velocyto, etc.)
  3. Cell Proportion analysis (scCODA, etc.)
  4. Cell signaling analysis (CellChat, CellPhoneDB, etc.)

You do not need to reproduce entire panels of figures: it is enough to reproduce a single plot / figure that represents one of the major analyses above.

Figures

  • Produce one figure per analysis that represents the most interesting results

Discussion

  • Write a paragraph per analysis that highlights the top results discovered and includes at least one citation per analysis that represents a potential interpretation or hypothesis from the results.

Analysis of your choice that was not performed in the publication

For the last part of the project, you will choose one additional analysis to perform on the data. Whatever analysis you choose must not have been already performed by the publication and must be one of the main analyses listed above or substantial enough as determined by me (i.e. a simple plot is not enough)

You may choose a tool or algorithm with the same constraints as before (it must have a documented code base and an associated publication). You may choose any of the analyses listed above. Please contact me if you are not sure if an analysis is substantial enough.

Figures

  • Create a figure or figure(s) that display the most pertinent results from the analysis

Discussion

  • Briefly remark on the results and why they may be interesting. Provide at least one citation from the literature that represents a potential hypothesis or point of interest from the results you show.

Final Project Grading

The requirements for the final project are:

  1. A single jupyter notebook or Rmarkdown that performs the above tasks, creates the required plots / visualizations and addresses the listed questions

The responses to the questions should be written in the style of a scientific publication, but you will be graded on the content. Each question can be answered in a short paragraph so we encourage you to be brief.