Team mitten_TDC19

DREAM Tumor Deconvolution Challenge, team mitten_TDC19 from Michigan State University, East Lansing, Michigan:

• Dr. Sergii Domanskyi ([email protected])
• Thomas Bertus ([email protected])
• Prof. Carlo Piermarocchi ([email protected])

Method description

This repository contains source code from synapse.org: mitten_TDC19 and documentation on how to run the scripts.

The method writeup can be found on synapse Wiki page of our team.

Here we provide pre-trained models in form of the signature matrices for Submissions 1, 2, 3 of the Final Round of the challenge for both coarse-grained and fine-grained subchallenges. The spreadsheet files are in signature directory of this repository. The name of each of the 6 signature matrices contains subchallenge identifier and final round number:

• File coarse_final_round_2.xlsx contains all markers from coarse_final_round_1.xlsx plus 63 additional markers for CD4.T.cells and 5 additional markers for CD8.T.cells. File coarse_final_round_3.xlsx is identical to coarse_final_round_1.xlsx.

• File fine_final_round_2.xlsx contains all markers from fine_final_round_1.xlsx plus 150 additional markers for myeloid.dendritic.cells and 9 additional markers for memory.CD4.T.cells. File fine_final_round_3.xlsx contains all markers from fine_final_round_1.xlsx plus 5 additional markers for myeloid.dendritic.cells and 9 additional markers for memory.CD4.T.cells.

Installation

Pre-Installation Requirement: to use this software on any platform you need Python version 3.7 or higher. Install packages numpy, pandas, scipy, xlrd, matplotlib, mygene, sklearn used in the scripts.

To clone this repository run:

git clone https://github.com/sdomanskyi/mitten_TDC19.git

Typical install time

Installation of the packages numpy, pandas, scipy, xlrd, matplotlib, mygene, sklearn on a normal desktop computer takes about 5 minutes. Installation of this git repository takes a few seconds.

Usage

Run python run_model.py --help to see description of all command line options.

Note: This repository contains source code that was used in Submissions 1, 2, 3 of the Final Round of the challenge, as well as the Leaderboard Rounds. The parameters exposed in the script run_model.py constitute the minimal set to reproduce the Final Rounds submissions and process any new user-provided data.

Input file input.csv should contain at least 3 columns as described below. Every row in the input file corresponds to a dataset:

Column	Description
dataset.name	Unique identifier of the dataset
scale	Descriptor of how the gene expression data was scaled, e.g. linear, log2
hugo.expr.file	Name of the file with normalized gene expression and gene symbol gene identifiers

Coarse-grained model run for Final Round 1:

python run_model.py --input=input/input.csv --model-level=coarse --signature-matrix=signature/coarse_final_round_1.xlsx

Fine-grained model run for Final Round 1:

python run_model.py --input=input/input.csv --model-level=fine --signature-matrix=signature/fine_final_round_1.xlsx

Coarse-grained model run for Final Round 2:

python run_model.py --input=input/input.csv --model-level=coarse --signature-matrix=signature/coarse_final_round_2.xlsx

Fine-grained model run for Final Round 2:

python run_model.py --input=input/input.csv --model-level=fine --signature-matrix=signature/fine_final_round_2.xlsx

Coarse-grained model run for Final Round 3:

python run_model.py --number-hvg=15 --input=input/input.csv --model-level=coarse --signature-matrix=signature/coarse_final_round_3.xlsx

Fine-grained model run for Final Round 3:

python run_model.py --number-hvg=15 --input=input/input.csv --model-level=fine --signature-matrix=signature/fine_final_round_3.xlsx

Note: In Final Round 3 we limited the number of highly variable genes to 15 by passing parameter --number-hvg=15.

Output file(s) is generated in the directory output. Below is the output file from the coarse-grained model run. Prediction is relative abundance score of a given sample for a given cell type.

dataset.name	sample.id	cell.type	prediction
FIAS5	S1	B.cells	2.225265108
FIAS5	S1	NK.cells	1.465895084
FIAS5	S1	neutrophils	2.135799171
FIAS5	S1	monocytic.lineage	2.449364958
FIAS5	S1	CD4.T.cells	0.606665336
FIAS5	S1	CD8.T.cells	0.305858142
FIAS5	S1	fibroblasts	0.657552838
FIAS5	S1	endothelial.cells	0.80516959
FIAS5	S2	B.cells	2.200313867
trimmed

Expected run time

It took approximately 1 minute to run each of the 6 scenarios detailed above.

Licensing

This software is released under an MIT License. Please also consult the file LICENSE in this repository regarding Licensing information for use of external associated content.

Funding

This work was supported by the National Institutes of Health, Grant No. R01GM122085.

Acknowledgements

We acknowledge discussions with Prof. Giovanni Paternostro and Alex Hakansson.