scRna-seq analysis & cell type annotation for SCPCP000018 {Osteosarcoma}

[ermismd]

Proposed analysis

The goal of this project is analyzing the single-cell RNA sequencing data from the 11 osteosarcoma samples in the SCPCP000018 dataset. Specifically, the aims are to:

1. Identify and annotate the distinct cell populations present in these samples.

2. Try refining any existing prior annotations, if available, to ensure more accurate cell-type assignments.

3. Try to correctly differentiate between normal and tumor-derived cell populations.

Scientific goals

The primary scientific objective of this analysis is to accurately delineate and annotate the cellular composition of the 11 osteosarcoma samples, leveraging the transcriptomic signatures to define each distinct cell population and refine existing labels where available. A secondary objective will be to distinguish non-malignant from malignant tumor cells, thereby characterizing the cellular heterogeneity and microenvironmental interactions that drive osteosarcoma.

Methods or approach

The analysis will follow a standard scRNA-seq pipeline in Python with scanpy:

1. Use Scrublet on raw counts to flag and remove likely cell doublets
2. Standard Data preprocessing:
   • Filter out cells with excessively low or high gene counts
   • Remove cells with a high fraction of mitochondrial & ribosomal transcripts
   • Filter out genes expressed in very few cells
3. Normalize counts and log-transform
4. Identify highly variable genes (HVGs) for downstream analyses
5. Get reference data for established cell type-specific marker genes from scientific literature and PanglaoDB
6. Dimensionality reduction with PCA using elbow method to inspect the variance explained and choose an appropriate number of PCs
7. Run Leiden algorithm at one or more resolutions to define clusters and plot UMAP to visualize clusters
8. Differential gene expression for each Leiden cluster to identify candidate marker genes
   • Use these markers together with the curated lists from the literature or PanglaoDB to guide annotation
9. Annotate clusters based on their expressed marker genes
10. Compare annotated clusters against any existing labels
11. For cancer vs normal inference apply inferCNV (infercnvpy) to detect large-scale copy-number changes and classify cells as likely malignant (aneuploid) versus non-malignant

Existing modules

N/A

Input data

The input data will be the scRNA-seq data from the Group ID SCPCAB0025 and with Project ID SCPCP000018.

Scientific literature

Overview of single cell analysis best practices: https://www.sc-best-practices.org/preamble.html

Osteosarcoma:

Ahmad Al Shihabi et all,The landscape of drug sensitivity and resistance in sarcoma,Cell Stem Cell,
Volume 31, Issue 10,2024,Pages 1524-1542.e4,ISSN 1934-5909, https://doi.org/10.1016/j.stem.2024.08.010.
(https://www.sciencedirect.com/science/article/pii/S1934590924002960)

Zhou, Y., Yang, D., Yang, Q. et al. Single-cell RNA landscape of intratumoral heterogeneity and immunosuppressive microenvironment in advanced osteosarcoma. Nat Commun 11, 6322 (2020). https://doi.org/10.1038/s41467-020-20059-6

Thomas DD, Lacinski RA, Lindsey BA. Single-cell RNA-seq reveals intratumoral heterogeneity in osteosarcoma patients: A review. J Bone Oncol. 2023 Mar 20;39:100475. doi: 10.1016/j.jbo.2023.100475. PMID: 37034356; PMCID: PMC10074210.

Liu W, Hu H, Shao Z, Lv X, Zhang Z, Deng X, Song Q, Han Y, Guo T, Xiong L, Wang B, Zhang Y. Characterizing the tumor microenvironment at the single-cell level reveals a novel immune evasion mechanism in osteosarcoma. Bone Res. 2023 Jan 3;11(1):4. doi: 10.1038/s41413-022-00237-6. PMID: 36596773; PMCID: PMC9810605.

He XY, Que LY, Yang F, Feng Y, Ren D, Song X. Single-cell transcriptional profiling in osteosarcoma and the effect of neoadjuvant chemotherapy on the tumor microenvironment. J Bone Oncol. 2024 May 8;46:100604. doi: 10.1016/j.jbo.2024.100604. PMID: 38765702; PMCID: PMC11101886.

Evola, Francesco R et al. “Biomarkers of Osteosarcoma, Chondrosarcoma, and Ewing Sarcoma.” Frontiers in pharmacology vol. 8 150. 7 Apr. 2017, doi:10.3389/fphar.2017.00150

scverse: https://scverse.org/

Other details

Timeline: August-October 2025

All data processing and analyses will be carried out on UT Southwestern’s BIOHPC high-performance computing cluster

[Jen-OMalley]

Hi @ermismd. I'm Jen, the Scientific Community Manager at the Data Lab. Thank you for interest in the OpenScPCA project! Our team is currently reviewing your proposed ideas, and we look forward to discussing more with you soon.

In the meantime, here is the contributor form. Filling out this form ensures you have agreed to the OpenScPCA terms and conditions and other policies. Once we receive this, we will get back to you here with any questions and/or next steps within three business days. We'll also start setting up an AWS account for you.

We look forward to hearing more about you plans!

[jaclyn-taroni]

Hi @ermismd, we've set up an AWS account for you. You should receive an email to complete the setup. Here is our documentation for joining IAM Identity Center.

[ermismd]

Awesome, thank you!

[allyhawkins]

Hi @ermismd, I'm Ally, 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 comments and questions about the proposal

Overall, I think your proposal sounds reasonable and is in line with a lot of the ways we have been thinking of cell type annotation. I have a few comments I wanted to make about the specific approach you are planning.

• You mention planning to get "reference data for established cell type-specific marker genes from scientific literature and PanglaoDB" in your proposal. All of the samples in ScPCA have already been annotated with CellAssign using a disease-appropriate marker gene reference from PanglaoDB. However, PanglaoDB is only helpful in annotating normal cell types and does not contain marker gene information about tumors. Because of that, I would encourage you to focus your marker gene search on those that will be helpful in identifying tumor cells or tumor cell states and cell types not represented in PanglaoDB. The hope is that these annotations will improve on existing annotations rather than recapitulate what we have already done.
• You also mention using Leiden clustering and then identifying candidate marker genes for each cluster to guide annotation. Can you expand on this point? How do you plan to annotate each cluster and then validate these assignments?
  • Also, note that if you are going to rely on clustering, we ask that you spend time validating and optimizing clustering parameters. You can find more information below on some resources we provide to help you do that.

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). 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, normalization, selection of highly variable genes, and dimensionality reduction.
  • 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!
• Since you plan to run InferCNV, we do have gene order files available for you to use. 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 let me know if I can clarify anything. I'm looking forward to chatting more about your proposal and helping you get started with contributions!

[ermismd]

Hi @allyhawkins,

Thank you so much for your feedback, and apologies for the delay in getting back to you.

Regarding your comments

My initial thought was to start with normal tissue markers to get a baseline for annotation, and then bring in tumor-specific markers from the literature to distinguish malignant cells. My main focus, however, will be on tumor marker genes. After assigning annotations, I also plan to run tools like InferCNV and CopyKAT to cross-check and validate the tumor-related clusters.

Clustering approach

For clustering, what I have in mind is to test different resolutions, numbers of neighbors, and PCs, especially when we already have an idea of the expected cell types, and then use DEGs within each cluster to see if they align with known marker genes.

For validation, I could combine several strategies:

• Internal cluster validation metrics (ex, Silhouette Score),
• Functional enrichment analysis of cluster-specific genes to see if they align with expected pathways or processes,
• Tools such as pyclustertree as well as Scedar, for evaluating cluster stability and structure.

On general contributing

I’ve already downloaded some of the processed samples and noted that preprocessing/QC and doublet detection have been done. I’ll use those results directly, and if I feel that alternative filtering is needed, I’ll provide clear rationale and evidence for the change. I also appreciate the guidance on InferCNV gene order files and will make sure to use those resources.

Question about the procedure

One quick question about the process: once I analyze a sample, should I upload the analysis script along with a short documentation/markdown file describing the steps and findings, and then wait for feedback before moving on to the next sample? I believe this is mentioned in the guidelines, but I wanted to confirm.

Thanks again for all your help and looking forward to moving ahead with this project.

Best,
Ermis

[allyhawkins]

Hi @ermismd, thanks for your reply! I'll reply to some of your comments directly below:

My initial thought was to start with normal tissue markers to get a baseline for annotation, and then bring in tumor-specific markers from the literature to distinguish malignant cells.

We do provide some initial cell type annotation by running both SingleR and CellAssign in the processed objects and identifying the consensus of the cell type annotation between those two methods. You can find these results in the consensus_celltype_annotation column of the cell metadata in the processed objects. We have found that this annotation does a pretty good job of identifying annotations for non-malignant cells.

However, we are limited by the references we use and in general those annotations can be quite broad, so it's possible we are missing some normal cells and could improve on the granularity of those annotations. I would encourage you to use those annotations as a guide and compare any annotations you identify through alternate methods to those existing annotations.

We also have curated a list of marker genes using CellMarker2.0 that could be useful in helping to validate normal cell type annotations that might be a good resource for you when it comes to identifying normal tissue markers. You can find that in the cell-type-consensus module here.

I still am not totally sure what you mean when you say you plan to use marker genes to get a baseline for normal annotation. Are you going to look at expression of these markers across all cells or is there another approach you had in mind? If so, how do you plan to identify a threshold to use for assigning the annotation?

My main focus, however, will be on tumor marker genes. After assigning annotations, I also plan to run tools like InferCNV and CopyKAT to cross-check and validate the tumor-related clusters.

I totally agree about running CNV analysis, but I just wanted to 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.

For clustering, what I have in mind is to test different resolutions, numbers of neighbors, and PCs, especially when we already have an idea of the expected cell types, and then use DEGs within each cluster to see if they align with known marker genes.

For validation, I could combine several strategies:

• Internal cluster validation metrics (ex, Silhouette Score),
• Functional enrichment analysis of cluster-specific genes to see if they align with expected pathways or processes,
• Tools such as pyclustertree as well as Scedar, for evaluating cluster stability and structure.

This sounds like a great approach!

I would encourage you to check out our hello-clusters module for some information on how you can easily calculate some of these metrics using our existing rOpenScPCA package. This should hopefully simplify some of the code you need when doing this! This package includes things like calculating silhouette width and cluster stability.

One quick question about the process: once I analyze a sample, should I upload the analysis script along with a short documentation/markdown file describing the steps and findings, and then wait for feedback before moving on to the next sample? I believe this is mentioned in the guidelines, but I wanted to confirm.

As you work through this process your code will be peer reviewed, yes! It is helpful to turn your work into smaller pieces so that it can make review easier and overall faster. For each "unit" of work or specific analysis you plan to complete, you should file an issue and then file a PR with the actual code and associated documentation.

That being said, you don't need to analyze each sample separately! I would encourage you to think of ways to break up your analysis plan so that you aren't duplicating work across each sample. You only need to write code to run inferCNV or evaluate expression of specific marker genes one time rather than separately for each sample. When initially developing your analysis it can be helpful to work with one or a handful of samples, but ultimately we want to have an analysis that we can run reproducibly across all samples.

Your first pull request should just be to initiate the analysis module following the instructions here. Given your analysis plan, your second PR should be geared towards annotating the normal cell types in these samples.

Please let me know if you have any more questions!

[ermismd]

Hi @allyhawkins,

Just to clarify on the normal annotation part, I meant looking at expression of the marker genes across clusters rather than across individual cells. Compare the relative expression patterns, and then validating against the consensus_celltype_annotations you mentioned.

I really appreciate the guidance (the cell-type-consensus resource will be really helpful,thank you for pointing me to it) and I’ll let you know if any other questions come up.

[allyhawkins]

@ermismd Sounds great! As a reminder, contributing to this project is an iterative process. We ask that you break the work up into smaller pieces and file pull requests as you go. Once you are ready to start your analysis, please follow the below 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.
• Initiate your module and file a pull request.

It sounds like your first step after creating your initial module will be to start working on clustering and optimizing parameters. Please let me know if you have any questions as you get started and looking forward to working with you on this analysis!