Expanding the coverage of spatial proteomics

Overview

This is the repository for the paper "Expanding the coverage of spatial proteomics" (Biorxiv). In thiswork, we first sought to develop a flexible approach for finding a small subset of markers and using them to predict the full-image expression pattern of the remaining markers, this enlarges the coverage of spatial proteomics. All code and intermediate results are available in this repository.

System and Hardware Requirements

Linux-based systems and NVIDIA graphics hardware with CUDA and cuDNN support are required. Our program is tested on Linux CentOS with CUDA 12.1.

Requirements

• python (tested on version 3.9.1)
• scikit-learn (tested on version 1.0.2)
• scipy (tested on version 1.10.0)
• matplotlib (tested on version 3.7.1)
• numpy (tested on version 1.25.1)
• tifffile (tested on version 2023.7.18)
• torch (tested on 1.9.1+cu111)
• networkx (tested on version 3.0)
• seaborn, tqdm (optional)

To reproduce the environment, use the following command:

pip install -r requirements.txt

Our experiments are conducted on the conda environment. If use conda to install the dependencies, use the following command:

conda env create -f environment.yml

Usage

To reproduce the results in the paper, open the Jupyter notebook files in the folder single or mp and run the cells. For example, to reproduce the single panel experiment results for spleen dataset, open the file panel_selection_spleen.ipynb under the folder single and run the cells.

The folder single is for single-panel experiments and the folder mp is for multi-panel experiments.

The following is the description of juptyernotebook files under single folder:

• panel_selection_spleen.ipynb is the single panel experiment for spleen dataset;
• panel_selection_lymph_node.ipynb is the single panel experiment for lymph node dataset;
• panel_selection_large_intestine.ipynb is the single panelexperiment for large intestine dataset;
• panel_selection_small_intestine.ipynb is the single panel experiment for small intestine dataset;
• panel_selection_pancreas.ipynb is the single panel experiment for pancreas dataset;

The following is the description of juptyernotebook files under mp folder:

• panel_selection_li is the multi-panel experiment for large intestine dataset;
• panel_selection_si is the multi-panel experiment for small intestine dataset;

Example

To reproduce the result of single panel experiment for spleen dataset:

• Download the data from HuBMAP. The image names can be found under the folder csv. train_path_SP.csv contains the training image names, val_path_SP.csv contains the validation image names, test_path_SP.csv contains the test image names. The image_address_spleen.txt stores the image addresses to download. Run the following command to download the data to the folder 'data/spleen':

cd data
wget -i image_address_spleen.txt -P spleen

• Find the jupyternotebook file panel_selection_spleen.ipynb under the folder single and run the cells. It will do the following steps:
  • Initialize the graph;
  • Select the minimal predictive subset of markers rounds by rounds. In each round:
    • Add the marker that has the highest predictive power for the remaining markers;
    • Train the model with the selected markers;
  • Evaluate the model.

It may take days to run the whole experiment. In each round, the running time for training a model will take about 4 hours to on a single GPU. We provide the intermediate results in the folder single/spleen_output. The intermediate results can be used to reproduce the results figures and tables without needing to run all the training steps.

Datasets

All the data are relevant CODEX image data from The Human BioMolecular Atlas Program (HuBMAP). The image names can be found under the folder 'csv'. The IMC pancreas dataset can be downloaded from The Human Tumor Atlas Network (HTAN).

Here also provides an instruction to download data by 'download_data.md'.

Acknowlegements

The fnet code is modified from AllenCellModeling/pytorch_fnet. We thank their contribution for providing the open source software.

TO_DO: License