Exploratory analysis of drug resistance associated cell states in paired neuroblastoma cell lines (SCPCP000020)

[akyneal]

Proposed analysis

I'd like to do an exploratory analysis (EDA) on SCPCP000020. (The four paired drug sensitive neuroblastoma cell line samples looking at differences between sensitive and resistant cells within each patient pair.

This would be an exploratory work, not a confirming study. I am a product manager by background (M.S. in Data Science), not a single-cell biologist. I am doing this as a personal volunteer project to learn the field, contribute a small piece of work back to the community, and help raise awareness for pediatric cancer research in my network. I plan to work locally first and share the resulting notebook publicly when it's in a presentable state. Posting here for community feedback before I get too far down the path.

Scientific goals

Exploratory problem: When we compare drug sensitive neuroblastoma cells to their drug resistant counterparts derived from the same patient, what changes at the transcriptional level?

1. Characterize per sample QC and baseline cellular heterogeneity in the four libraries.
2. Score every cell against published marker gene sets and compare distributions across sensitive vs. resistant lines.
3. Identify the top differentially expressed genes between sensitive and resistant cells within each patient pair.

One thing i noticed; SMS pair has a primary tumor vs bone marrow metastasis has tissue origin difference on top of drug status which i am not sure if right way for comparisons.

Methods or approach

Standard EDA pipeline in Python Jupyter notebook; will work locally first

-Load all four processed AnnData from the Portal download.
-QC summary of all across libraries
-Focus on single AnnData with sample level metada.
-Visualize existing clusters from the ScPCA processing pipeline. (sanity check)
-Cell state scoring
-Cross pair consistency check possibly

I will frame cell type rather than cell line.

Existing modules

I am not aware of any existing OpenScPCA module focused on this.
This should consume the standard processed AnnData outputs from the ScPCA pipeline.

Input data

• SCPCP000020 processed AnnData files for the four libraries: SCPCL000944 (CHLA-15), SCPCL000945 (CHLA-20), SCPCL000946 (SMS-KAN), SCPCL000947 (SMS-KANR).
• single-cell_metadata.tsv

Scientific literature

I need to research this.

Other details

I would like to use my local laptop first with no AWS, GCP access.
I am hoping to do this within my bandwidth , 1 -2 weeks and share a notebook.

I just want to re-iterate my background. I am a product manager and I usually deal with Revenue Intelligence, Marketing Budget allocation, predictive scoring, personalization and etc... Hence i dont have the domain expertise for single-cell analysis.

I am interested to learn how to work with scientific medical data. (single-cell analysis in this case) and raise awareness within other Data Scientists / Researchers within my network.

I am open to feedback.

[jaclyn-taroni]

Hi @akyneal,

Thanks for this proposal and your interest in contributing to OpenScPCA!

To give you some high-level feedback on your approach:

• We attempt "reasonable defaults" for clustering that's part of the ScPCA processing pipeline, but we do not necessarily expect the parameters we used to be optimal for every dataset. If you are particularly interested in examining heterogeneity, I might suggest a deeper dive into clustering in one notebook. We have implemented functionality for evaluating clusters in R that's demonstrated in this module – https://github.com/AlexsLemonade/OpenScPCA-analysis/tree/main/analyses/hello-clusters – but I assume there is similar functionality in Python libraries designed for scRNA-seq data.

• One difference in cell states or lineages that is of interest in neuroblastoma is mesenchymal vs. adrenergic (for example, see: https://pmc.ncbi.nlm.nih.gov/articles/PMC10171398/), so I would look at the literature for adrenergic and mesenchymal marker gene sets.
• You can certainly do descriptive things like look at the distribution of gene set scores between samples, but I think the sample size is going to be a problem for formal differential expression testing (a limitation of this particular dataset).

We think the best way to get started is to create an analysis module as described in the docs (https://openscpca.readthedocs.io/en/latest/contributing-to-analyses/analysis-modules/creating-a-module/), which will create a skeleton structure. It'd be great to have you file a pull request establishing the module before you send one with any EDA notebooks to keep the pull request size manageable.

Thanks again for your interest! Please let us know if there's any way we can assist you with your contributions.

[akyneal]

Thanks @jaclyn-taroni for your detailed feedback. I will further;

• Drop the formal differential expression as the analysis given the sample size limitation and work towards descriptive comparisons of gene set score distribution instead.
• add a clustering deep dive in a dedicated notebook. I work with python so i will use that but use the R hello-cluster as a reference.
• Will also ground the cell state scoring to the PMC10171398.

I am going to follow the docs to create the analysis module skeleton and file that PR first. After that i will update with the EDA notebooks in a separate PR.

Will reach out via discussions if any questions.

Thanks again.