GMM-Demux 0.2.0.1

A multiplet removal tool for processing cell hashing data

GMM-Demux

A Gaussian Mixture Model based software for processing sample barcoding data (including both cell hashing and MULTI-seq data).

Below shows an example classification result. Orange dots are multi-sample multiplets.

Description

GMM-Demux removes Multi-Sample-Multiplets (MSMs) in a cell hashing dataset and estimates the fraction of Same-Sample-Multiplets (SSMs) and singlets in the remaining dataset. GMM-Demux also verifies if a putative cell type exists, or is it merely an artifact induced by multiplets.

Multiplet-induced fake cell types are called "phony cell types".

An example phony cell type in a CITE-seq dataset is provided in the picture below:

In the above figure, both CD3+CD19+ and CD4+CD8+ cell types are multiplet induced fake cell types.

Phony type clusters have large percentages of MSMs, as above figure shows. Both phony type clusters have large MSM percentages.

Percentages of MSMs are used as key features by GMM-Demux to classify GEM clusters.

Terminology

• Singlet: Droplet that contains a single cell.

• MSM: Multi-Sample Multiplet. A multiplet that contains cells from different samples in sample barcoding. MSMs can be identified by GMM-Demux.

• SSM: Same-Sample Multiplet. A multiplet that contains cells from the same sample in sample barcoding. SSMs cannot be separated from singlets by sample barcoding.

• SSD: A combined category of both SSMs and singlets.

• Pure type: a real cell type that exist in the tissue.

• Phony type: an artificial cell type that is an artifact produced by multiplets.

• Mixture type: a cluster of droplets in which there exist a large fraction of phony type droplets.

An illustration of above terminology is provided in the figure below:

Features

• Remove cell-hashing-identifiable multiplets (MSMs) from the dataset.
• Estimate the fraction of cell-hashing-unidentifiable multiplets (SSMs) in the remaining dataset (the RSSM percentage).
• Tests if a putative cell type is a pure (real) cell type or is it a phony cell type.

Example Dataset

• An example cell hashing data is provided in the example_input folder. It contains the per drop HTO count matrix of a 4-sample cell hashing library prep. The input folder has the same file format with the CellRanger v3 output.

Authors

Hongyi Xin, Qi Yan, Yale Jiang, Jiadi Luo, Carla Erb, Richard Duerr, Kong Chen* and Wei Chen*

Maintainer

Hongyi Xin

Requirement

GMM-Demux requires python3 (>3.5).

Install

GMM-Demux can be directly installed from PyPi. Or it can be built and installed locally.

Install GMM-Demux from PyPi.

pip3 install --user GMM_Demux

In some OS, the pip3 is linked to pip by default. For these OS, the installation command is simply:

pip install --user GMM_Demux

Check if pip3 is linked to pip with pip -V.

If choose to install from PyPi, it is unnecessary to download GMM-Demux from github. However, we still recommend downloading the example dataset to try out GMM-Demux.

Install GMM-Demux locally using setuptools and pip3.

You may choose to install it locally from the github repository. However, this is for advanced users only and support is not gauranteed.

cd <GMM-Demux dir>
python3 setup.py sdist bdist_wheel
pip3 install --user . 

Post installation processes

If this is the first time you install a python3 software through pip, make sure you add the pip binary folder to your PATH variable. Typically, the pip binary folder is located at ~/.local/bin.

The pip binary folder might locate at a different location if the user uses virtual enviroment. Pay attention to the pip installation output.

Here is an example installation output. The path of the pip binary folder is highlighted:

To temporarily add the pip binary folder, run the following command:

export PATH=~/.local/bin:$PATH

To permenantly add the pip library folder to your PATH variable, append the following line to your .bashrc file (assuming the user uses bash as the default shell).

PATH=~/.local/bin:$PATH

Content

The source code of GMM-Demux is supplied in the GMM_Demux folder.

An example cell hashing dataset is also provided, located in the example_input/outs/filtered_feature_bc_matrix folder.

An example set of hand-curated putative cell types are provided in the example_cell_types folder.

An example csv HTO file of the above cell hashing data is provided as the example_hto.csv file.

Usage

Case 1: Basic Usage, Remove MSMs

Once installed, GMM-Demux is directly accessible with the GMM-demux command.

GMM-demux <cell_hashing_path> <HTO_names>

<HTO_names> is a list of strings separated by ',' without whitespace. For example, there are four HTO tags in the example cell hashing dataset supplied in this repository. They are HTO_1, HTO_2, HTO_3, HTO_4. The <HTO_names> variable therefore is HTO_1,HTO_2,_HTO_3,HTO_4.

MSM-free droplets are stored in folder GMM_Demux_mtx under the current directory by default. The output path can also be specified through the -o flag.

Example Command

An example cell hashing data is provided in example_input. <HTO_names> can be obtained from the features.tsv file.

GMM-demux example_input/outs/filtered_feature_bc_matrix HTO_1,HTO_2,HTO_3,HTO_4

<HTO_names> are obtained from the features.tsv file. The feature.tsv file of the example cell hashing dataset is shown below.

Output

MSM-free droplets, in MTX format. The output has the same format with CellRanger 3.0 outputs. By default, the output is stored in SSD_mtx folder. The output location can be overwritten with -o flag.

Case 2: Compute the MSM and SSM rates

To compute the MSM and SSM rates, GMM-Demux requires the -u flag:

• -u SUMMARY, --summary SUMMARY Generate the statstic summary of the dataset. Requires an estimated total number of cells in the assay as input.

-u flag requires an additional <NUM_OF_CELL> argument, which is the estimated total count of cells in the single cell assay.

Example Command

GMM-demux example_input/outs/filtered_feature_bc_matrix HTO_1,HTO_2,HTO_3,HTO_4 -u 35685

Output

Below is an example report:

• RSSM denotes the percentage of SSM among the remaining SSDs (after removing all MSMs). RSSM measures the quality of the cell hashing dataset.

Case 3: Verify if a cell type exists

GMM-Demux verifies a putative cell type with the -e flag:

• -e EXAMINE, --examine EXAMINE Provide the cell list. Requires a file argument. Only executes if -u is set.

-e flag requires a file name, which stores the list of droplet barcodes of the putative cell type.

Example Command

GMM-demux example_input/outs/filtered_feature_bc_matrix HTO_1,HTO_2,HTO_3,HTO_4 -u 35685 -e example_cell_types/CD19+.txt
GMM-demux example_input/outs/filtered_feature_bc_matrix HTO_1,HTO_2,HTO_3,HTO_4 -u 35685 -e example_cell_types/Doublets/CD3+CD4+CD19+.txt

Output

An example output of a pure cell type:

An example output of a phony cell type:

Case 4: Use the csv file format as input, instead of the mtx format

Example Command

GMM-demux -c example_hto.csv HTO_1,HTO_2,HTO_3,HTO_4 -u 35685

Case 5: Extract droplets of specific HTO sample configurations

Extract droplets that are labeled with specific HTO(s), with the -x flag:

• -x EXTRACT, --extract EXTRACT Names of the HTO tag(s) to extract, separated by ','. Joint HTO samples are combined with '+'.

When -x is set, other functions of GMM-Demux will be turned off.

Case 5a: Extract a single HTO sample

Example Command

GMM-demux example_input/outs/filtered_feature_bc_matrix HTO_1,HTO_2,HTO_3,HTO_4 -x HTO_1

Case 5b: Extract a single HTO sample that are jointly defined by two HTO tags

Use + to specify the joint HTO tags.

Example Command

GMM-demux example_input/outs/filtered_feature_bc_matrix HTO_1,HTO_2,HTO_3,HTO_4 -x HTO_1+HTO_2

Case 5c: Extract multiple HTO samples

Use , to separate sample tags. Single tag samples can be merged with joint-tag samples.

Example Command

GMM-demux example_input/outs/filtered_feature_bc_matrix HTO_1,HTO_2,HTO_3,HTO_4 -x HTO3,HTO_1+HTO_2,HTO_1+HTO_4

Optional Arguments

• -h: show help information.
• -f FULL, --full FULL Generate the full classification report. Require a path argument.
• -s SIMPLIFIED, --simplified SIMPLIFIED Generate the simplified classification report. Require a path argument.
• -o OUTPUT, --output OUTPUT The path for storing the Same-Sample-Droplets (SSDs). SSDs are stored in mtx format. Requires a path argument. Default path: SSD_mtx.
• -r REPORT, --report REPORT Specify the file to store summary report. Require a file argument.
• -c CSV, --csv Take input in csv format, instead of mmx format.
• -s SKIP, --skip FULL_REPORT Load a full classification report and skip the mtx folder as input. Require a path argument.
• -a AMBIGUOUS, --ambiguous AMBIGUOUS The estimated chance of having a phony GEM getting included in a pure type GEM cluster by the clustering algorithm. Requires a float in (0, 1). Default value: 0.05. Only executes if -e executes.
• -t THRESHOLD, --threshold THRESHOLD Provide the confidence threshold value. Requires a float in (0,1). Default value: 0.8.

Parsing the Classification Output

There are two files in a classification output folder. A config file (ending with .config) and a classification file (ending with .csv).

The classification file contains the label of each droplet as well as the probability of the classification. The classification is represented with numbers which are explained in the config file.

Below shows the classification output of the example data:

Online Cell Hashing Experiment Planner

A GMM-Demux based online cell hashing experiment planner is publically accessible at here.

Citation

If you find this code useful in your research, please consider citing:

@article{xin2019sample,
  title={Sample demultiplexing, multiplet detection, experiment planning and novel cell type verification in single cell sequencing},
  author={Xin, Hongyi and Yan, Qi and Jiang, Yale and Lian, Qiuyu and Luo, Jiadi and Erb, Carla and Duerr, Richard and Chen, Kong and Chen, Wei},
  journal={bioRxiv},
  pages={828483},
  year={2019},
  publisher={Cold Spring Harbor Laboratory}
}

Acknowledgement

Special thank to Zhongli Xu for testing GMM-Demux!