Immune infiltrate quantification

We are looking to understand the current state-of-the-art and motivate new methods:

• Understand how existing methods extract representative marker genes or expression profiles from immune cell types, and their drawbacks
• Understand differences between microarry and RNAseq deconvolution
• Understand how existing methods perform deconvolution of an unknown mixture based on reference marker genes or expression signatures, and the drawbacks of these methods
• Show test cases where the existing methods fail
• Collect training/ground truth data from existing papers that we can use
• Prototype new deconvolution models.

See Summary.ipynb for an overview of the analysis. Here's also the latest project review.

(Summary.html is a snapshot of Summary.ipynb with better rendering of the table of contents than Github's rendering of Summary.ipynb. It's also saved into gh-pages and hosted at www.hammerlab.org/immune-infiltrate-explorations/Summary.html. The table of contents is made by the TOC jupyter nbextension.)

Directories:

• curated_data/: data downloads. Some data is not in Git and is only in a Google bucket -- see README in curated_data.
• data_engineering/: some transformations of the data
• evaluate_existing_methods/: analysis to evaluate existing deconvolution methods
• hierarchical_models/: new hierarchical classification models
• behavior_of_rnaseq_data/: analysis of how RNA-seq data behaves in comparison to microarray data
• plots/: generated figures
• reference-extraction_writeup/reference_extraction.pdf: an analysis of the reference extraction -- linked from Summary notebook