editor_options
markdown wrap 72

MetastaticMelanomaSC

This repo focuses in the re-analysis of the dataset developed by Tirosh and colleagues in the paper "Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq" from 2016.

• The folder Rmarkdown contains the main report for this project: ProjectReport.Rmd. It contains the code to reproduce the data analysis and to generate the input feature tables for the supervised classification pipeline.

• Under the main folder the file ProjectPresentation.pdf contains the PDF version of the presentation for future references.

• The folder Data contains all the files downloaded from the original publications and the feature tables.

• The folder JupyterNB contains the jupyter notebook that can be used to run the supervised learning pipeline and calculate the SHAP values for the best trained model.

To facilitate data exploration I developed the following Shiny apps:

• Reactive document (sample dataset): https://mtello.shinyapps.io/reactive_doc/
• Static tSNE visualization (full dataset): https://mtello.shinyapps.io/tsneviz/

Below is the original project proposal:

Re-analysis proposal

In this study, Tirosh and colleagues sampled 18 melanoma tumors and performed single cell RNA-seq (scRNA-seq) profiling. They defined gene clusters using standard PCA to determine biological significance of tumor cells (1). Since most of their interpretation is based on clusters defined within the tumor cells, it is possible that the clusters do not take full advantage of the transcriptomic variability in the dataset. I aim to improve the understanding of gene relevance for tumor-specific functions by incorporating microenvironment transcriptomic profiles. Due to advances of interpretability tools for supervised regressors, it is possible to see inside of "black box" models to understand what features drive differences between conditions. For instance, Yap and colleagues 2021 utilized the SHAP framework to determine individual gene contribution for classifying bulk RNA-seq samples into different tissues (2). I intend to apply this framework in scRNA-seq to determine if there are genes with a more subtle role in separating tumor vs micro-environment cells.

Methodology plan: Train a gradient boosting model to classify cells between "tumor" or "micro-environment". Use the SHAP framework to determine which genes have the biggest role during classification. Compare similarities of "SHAP genes" with results from a differential expression analysis.

Hypothesis: If the supervised model identified genes relevant for the tumor or micro- environment classification, this should be reflected in the gene SHAP profile. If the genes have a significant role in separating micro-environment from tumor cells, I expect a significant overlap between differential expressed genes and "SHAP genes"

Contribution: This analysis explores the utility of supervised classifiers to draw biological meaningful insights that separate cells from the melanoma micro-environment and the actual tumor.

References:

1. Tirosh I, Izar B, Prakadan SM, Wadsworth MH, Treacy D, Trombetta JJ, et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science. 2016 Apr 8;352(6282):189--96.

2. Yap M, Johnston RL, Foley H, MacDonald S, Kondrashova O, Tran KA, et al. Verifying explainability of a deep learning tissue classifier trained on RNA-seq data. Sci Rep. 2021 Jan 29;11(1):2641.