Proposed Analysis for SCPCAB0024: Cell Type Annotation of Pediatric Osteosarcoma

[khoidnutsw]

Proposed analysis

This analysis proposes to perform comprehensive cell type annotation for the single-cell RNA sequencing dataset associated with Group ID SCPCAB0024 (Project ID SCPCP000017), which focuses on pediatric osteosarcoma. The goal is to identify and characterize the various cell populations present within the tumor microenvironment and circulating tumor cells, providing a detailed cellular atlas, with a specific focus on refining cell-level annotation and distinguishing cancer cells from normal cells.

Scientific goals

The scientific goals of this analysis are directly aligned with the project's objectives as outlined in its abstract:

1. Describe the tumor and immune microenvironment: Accurately identify and quantify the diverse cell types (e.g., malignant osteosarcoma cells, various immune cell subsets, stromal cells) within the primary and metastatic osteosarcoma samples.

2. Analyze individual circulating tumor cells (CTCs): Characterize the phenotypic and potential phylogenetic relationship of CTCs to primary and metastatic disease by assigning their cell types and comparing their profiles to those of tumor cells from solid tissues.

3. Correlate gene expression patterns with clinical behavior: Lay the groundwork for future analyses that correlate the identified cell types and their gene expression patterns with clinical variables such as the presence of metastasis and response to therapy, ultimately aiming to discover biomarkers predictive of clinical behavior.

4. Distinguish cancer from normal cells at the single-cell level: Develop and apply methods to accurately identify malignant osteosarcoma cells and differentiate them from non-malignant cells within the tumor microenvironment and circulation.

Methods or approach

The cell type annotation will follow a standard single-cell RNA sequencing analysis workflow, adapted for robust and accurate cell type identification, with additional steps for cancer/normal cell detection:

1. Data Preprocessing: Quality control, normalization, and dimensionality reduction (e.g., PCA, UMAP) of the raw gene expression data.

2. Clustering: Unsupervised clustering algorithms (e.g., Leiden or Louvain clustering) will be applied to group cells with similar gene expression profiles.

3. Marker Gene Identification: Differential gene expression analysis will be performed for each cluster to identify unique marker genes.

4. Cell Type Annotation:

• Manual Annotation: Clusters will be manually annotated based on established cell type-specific marker genes from scientific literature and publicly available databases (e.g., CellMarker, Human Cell Atlas).

• Reference-based Annotation (if applicable): Utilize computational tools (e.g., Seurat's FindTransferAnchors and TransferData, SingleR, Azimuth) to integrate and annotate cells using well-curated reference datasets of known cell types, particularly for immune cell populations.

5. Doublet Detection and Removal: Implement methods to identify and remove potential doublet cells to ensure accurate cell type assignments.

6. Cancer/Normal Cell Detection:

• Copy Number Variation (CNV) Inference: Infer large-scale chromosomal CNVs from single-cell RNA-seq data (e.g., using inferCNV, CopyKAT) to identify aneuploid cells characteristic of cancer.

• Differential Expression Analysis: Compare gene expression profiles between putative tumor clusters and normal cell clusters (e.g., fibroblasts, epithelial cells from normal tissue if available) to identify cancer-specific gene signatures.

• Integration with Clinical/Genomic Data: If available, integrate information on known genomic alterations in osteosarcoma to support the identification of malignant cells.

7. Pathway Analysis: Perform pathway enrichment analysis on identified cancer cell clusters to understand underlying biological processes.

8. Refined Cell-Level Annotation: Based on the combined results of marker gene analysis, reference-based annotation, and cancer/normal cell detection, refine the cell type assignments to a granular level, distinguishing between malignant and non-malignant cell types.

9. Visualization: Generate high-quality visualizations (e.g., UMAP/t-SNE plots colored by cell type and cancer/normal status, heatmaps of marker genes, CNV profiles) to represent the identified cell populations.

10. Report Generation: A comprehensive report detailing the methods, results, and visualizations will be generated, suitable for contribution to the OpenScPCA-analysis repository.

Existing modules

This analysis will consume the processed single-cell RNA-seq data from the SCPCP000017 project. It is anticipated that the output of this analysis (cell type assignments, including cancer/normal status) will serve as input for downstream analyses, such as differential expression analysis between disease states or trajectory inference, potentially relating to existing or future modules focused on biological interpretation.

Input data

The primary input data will be the single-cell RNA-seq data (gene expression matrices and associated metadata) for the 28 osteosarcoma samples within Project ID SCPCP000017 (Group ID SCPCAB0024). This includes both primary tumor and circulating tumor cell data, as described in the project details.

Scientific literature

Relevant scientific literature on osteosarcoma biology, single-cell RNA sequencing analysis, and cell type annotation methodologies will be consulted.

Other details

No response

[Jen-OMalley]

Hi @khoidnutsw. I'm Jen, the Scientific Community Manager at the Data Lab. Thank you for sharing your proposed analysis! Our team will be reviewing your submission and will follow up with next steps soon. You can expect to hear from us by next Wednesday.

We will also be setting up an AWS account for you. Once we do, you should receive an email to finish setting up your account. Here are instructions for setting up AWS. I'll reach out again when you should be expecting to see this.

In the meantime, please let me know if you have any questions about OpenScPCA. We look forward to working together!

[Jen-OMalley]

Hi @khoidnutsw. You should have received the email to finish setting up your AWS account yesterday. Please reach out if you did not receive it or if you have any questions!

[khoidnutsw]

Hello Jen, could you resend the AWS invitation email? Thank you

[jaclyn-taroni]

Hi @khoidnutsw, I've sent an email to reset your user password. I hope that does the trick. Please check your spam folder just in case.

[sjspielman]

Hi @khoidnutsw, I'm Stephanie, one of the Data Scientists with the Childhood Cancer Data Lab here to help you get started on your analysis! To begin, I'd like to discuss some aspects of your proposed analysis to both give you some tips on working on your module, and to learn more about your planned approach.

Specific questions about the proposal

First, I would like to ask some follow-up questions about your proposal. Overall, it seems like you have laid out a very strong concept for a general cell type annotation pipeline, but it is not very specific to this data and/or the exact tasks you plan to carry out here. It would be great to learn more about what you're thinking for these particular samples :)

For example, here are some of the specific questions I have, to give you a sense of what level of detail we're looking for before you get started.

• You list several cluster algorithms (Louvain and Leiden); which do you plan to use and how will you decide? Again, please bear in mind that cluster validation will be important here!
• Under the cell type annotation section:
  • For "manual annotation" - which marker genes do you plan to use, and what are their literature sources? Note also that module reproducibility is critical for the OpenScPCA project, so any manual steps will still need to be reproducible with code and well-documented. Also, note that we have already used PanglaoDB marker genes as part of our automated cell typing, so we'd like to see annotations here that can potentially improve on those rather than recapitulate
  • For "reference-based annotation," you indicate that it might not be applicable, and you listed several methods. How will you decide if it is applicable? Which of these methods will you use and why? You also write that you may integrate data before this step - can you tell me more about the motivation for integration vs annotating each sample separately, and further what methods you would use for integration?

Please feel free to take some time to think about the specific analysis steps you would like to take here, since this will both help you to structure your analysis, and it will help us to review your code and support you along the way! As an example of this, one important part of OpenScPCA-analysis contributing is scoping pull requests to ensure manageable units of work as you build up the module. Having a detailed scientific plan in place will help you to plan your work and scope your PRs to have a much smoother experience overall. Let's continue the discussion here as you refine your analysis plan, happy to chat anytime! :)

General information about contributing

In addition, I wanted to offer some general information about contributing to OpenScPCA as it relates to some items in your proposal:

• Rather than performing your own pre-processing, we ask that you start your analysis with the processed objects available on the ScPCA Portal (_processed.rds or _processed_rna.h5ad; likely you'll want the RDS files based on your proposed methods). This helps keep analyses across modules uniform and transparent, and ensures all analyses have the same starting point. These processed objects have already undergone removal of empty droplets, filtering of low quality cells, and normalization.
  • Generally speaking, if you feel you need to use different filtering, normalization, etc. methods than are already included in the project, please provide your rationale and evidence that supports the alternative method(s).
  • You can see the documentation on the ScPCA Portal for more information about how these objects were processed. When you are ready to get started, you can download the data using the download-data.py script in the repository, which will download the processed objects by default.
• As part of OpenScPCA we have also annotated doublets in all samples using the steps in the doublet-detection module, which uses scDblFinder to annotate doublets. This means you can filter doublets directly using the scDblFinder results we provide rather than having to run this code yourself.
  • Again, we prefer that you use these results for detecting and filtering doublets. If you prefer a different approach, please provide your rationale and evidence that supports the alternative method(s).
  • For more information about obtaining these files and their content, see this announcement: <Now available: Doublet detection results across ScPCA datasets #684.
• Since you plan to incorporate clustering into your annotation pipeline, I wanted to let you know that we have an R package called rOpenScPCA for your convenience to support common analyses in the project.
  • This package includes functionality for calculating clusters and evaluating their quality. Complete usage examples for rOpenScPCA clustering functions are available in the hello-clusters analysis module.
  • In general, if you want to use clusters as the basis for a set of annotations, it's going to be important to validate those clusters, so we hope this package can help you with that process!
• At least one of your proposed steps may involve using Seurat objects, e.g. Azimuth label transfer. If you indeed plan to use Seurat, it might be useful to know that we also have converted all processed objects to Seurat v5 format using the steps in the seurat-conversion module. This means you can directly start your analysis from a Seurat object instead of a SingleCellExperiment object.
  • For more information about obtaining these objects and their content, see this announcement: Seurat objects now available in results buckets #996.
• For CNV analyses, please note that we've experienced some issues running copyKAT on some of the test datasets we have set up for continuous integration testing in this project, and we have observed that copyKAT inferences are not exactly reproducible. Because of these issues, we do have a preference for inferCNV over copyKAT as a method to detect CNV events.
  • Along those lines, note that we have gene order files available for use with inferCNV. You can learn more about those files and how to obtain them from this announcement: InferCNV gene order files now available in results bucket #1227

---

Thanks for your interest in OpenScPCA, and looking forward to chatting more about your proposal! Let me know if I can clarify anything so far too :)

[khoidnutsw]

Hi Stephanie,

Thanks so much for the detailed feedback and for helping me get started! I appreciate the guidance on streamlining the analysis and leveraging the existing OpenScPCA resources.

Regarding your questions about the proposal, I clarify my analysis intention as below:

• Clustering Algorithm: I plan to use Leiden clustering. My rationale is that Leiden generally produces more connected communities than Louvain, which can be beneficial for biological data where cell populations often form continuous spectra. I am thinking of using a metric to evaluate how well the clusters are separated (Silhouette Score), also using marker genes to adjust the algorithm.

• Cell Type Annotation:

1. Manual Annotation: I will use well-established marker genes from recent osteosarcoma-specific single-cell RNA-seq literature to identify malignant cells, various immune subsets (e.g., T cells, B cells, macrophages, neutrophils), and stromal cells (e.g., fibroblasts, endothelial cells)
2. Reference-based Annotation: I will use several automated tools such as SingleR or scmap, etc. My goal is to use at least 2 tools with 2 different database.
3. Integration: The motivation for integration, specifically using Harmony or Seurat's IntegrateData (depending on the dataset characteristics), is to correct for batch effects that may arise from different patients or processing batches, allowing for more accurate clustering and cell type identification across the entire dataset. It also helps to detect rare cell types. I also would then perform annotation on this integrated dataset.

• Cancer/Normal Cell Detection: I will prioritize inferCNV for Copy Number Variation inference as you recommended. I'm glad to hear the gene order files are available, which will be very helpful. This will be a critical step in differentiating malignant osteosarcoma cells from non-malignant cells. I will also incorporate differential expression analysis between putative tumor clusters and normal cells within the tumor microenvironment to identify cancer-specific gene signatures.

[sjspielman]

Hi @khoidnutsw,

Thanks for your reply!

I do have a couple followups below - these items are very important for you to nail down before proceeding with a given analysis stage and writing any code/documentation/reference files. You can either to think about these questions now, or think about them more carefully as you reach that stage of analysis:

1. First, to clarify, what exactly do you mean that you will "use marker genes to adjust the [clustering] algorithm"? I'm not sure what this is referring to.
2. For any steps using marker genes, those marker genes will need to be decided on before the analysis begins so that you can add them as reference files to your analysis module. For those PRs to be approved during code review, you will need to have clear sources and justification for the proposed marker genes sets. I recommend taking the time before you begin your analysis to identify what these will be.
3. For reference based annotation, it sounds like you still have not firmly decided what methods or references to use. Again, I encourage you figure this out as soon as possible to ensure a smooth contribution process.
4. For integration, it again sounds like you have not chosen the method you plan to use here. But, I would suggest to be very careful which methods are suitable for use on an integrated vs. single-sample dataset, both in terms of assumptions the methods make and the runtime for those methods. Annotating on an integrated object may be computationally prohibitive.

We don't have to work all this out now, but having a more concrete plan in place before beginning a step will be critical for having a smooth contribution process, in particular during code review. So, this discussion is always available for you to come back and chat with us about your plans and ideas as you are working on your analysis!

Once you feel you are ready to start your analysis, please follow these steps to start contributing to the project:

• Follow the technical setup instructions found in the OpenScPCA documentation.
• File an issue to track the initiation of your analysis module.
  • As you proceed with your analysis, you can open additional issues corresponding to the individual steps/PRs you will make
  • Again, please bear in mind this documentation on scoping your pull requests as you plan your work.
• Initiate your module and file a pull request.
  • Please name your analysis module cell-type-osteosarcoma-17. This indicates that you are performing cell type annotation on osteo samples from ScPCA project SCPCP000017.

If you have any questions during this process, let me know!