scPROTEIN (single-cell PROTeomics EmbeddINg) is a deep contrastive learning framework for single-cell proteomics embedding.
The advance of single-cell proteomics sequencing technology sheds light on the research in revealing the protein-protein interactions, posttranslational modifications, and proteoform dynamics of proteins in a cell. However, the uncertainty estimation for peptide quantification, data missingness, batch effects and high noise hinder the analysis of single-cell proteomic data. It is important to solve this set of tangled problems together, but the existing methods tailored for single-cell transcriptomes cannot fully address this task. Here, we proposed a novel versatile framework designed for single-cell proteomics data analysis called scPROTEIN, which consists of peptide uncertainty estimation based on a multi-task heteroscedastic regression model and cell embedding generation based on graph contrastive learning. scPROTEIN can estimate the uncertainty of peptide quantification, denoise protein data, remove batch effects and encode single-cell proteomic-specific embeddings in a unified framework. We demonstrate that scPROTEIN is efficient for cell clustering, batch correction, cell type annotation, clinical analysis, and spatially resolved proteomic data exploration.
For more information, please refer to https://www.biorxiv.org/content/10.1101/2022.12.14.520366v1
A csv file in the following format is needed for scPROTEIN learning from stage 1:
| Protein | Peptide | Cell 0 | Cell 1 | Cell 2 | Cell 3 | Cell 4 | Cell 5 | Cell 6 |
|---|---|---|---|---|---|---|---|---|
| P08865 | LLVVTDPR_2 | 0.215943903 | 1.825849332 | 0.17106779 | 0.090752671 | 0.633329732 | -0.044091136 | NA |
| P26447 | RTDEAAFQK_3 | 1.873431237 | 1.425136257 | 2.354956659 | 1.373487482 | 1.724188343 | 0.828024968 | 0.511722654 |
| P26447 | LNKSELK_3 | NA | NA | NA | NA | NA | -0.164518259 | -0.765802428 |
| Q00610 | LLYNNVSNFGR_2 | -0.452033525 | NA | NA | -0.211513228 | -0.573607252 | -0.593867542 | NA |
| P05120 | LNGLYPFR_2 | NA | NA | 0.245379509 | 0.923845132 | 0.300612918 | NA | NA |
"Protein" represents the protein name and "Peptide" denotes the corresponding constituting peptide sequence(s). The columns "Cell 0","Cell 1"... are the protein data in each cell. NA is the missing value. If datasets are provided directly from protein-level (without "Peptide" column), scPROTEIN can start from stage 2.
The documentation which elucidates the functions of scPROTEIN is provided.
Recomended usage procedure is as follows.
1.Installation
The running environment of scPROTEIN can be installed from docker-hub repository:
- Pull the Docker image from docker-hub
docker pull nkuweili/scprotein:latest
- Run the Docker image (GPU is needed)
docker run --name scprotein --gpus all -it --rm nkuweili/scprotein:latest /bin/bash
- Download this repository (This usually takes 15 seconds on a normal desktop computer)
# If you encounter issue about proxy when running "git clone" in the Docker environment on your own device, you can first execute the following command before running "git clone":
# git config --global http.proxy ""
git clone https://github.com/TencentAILabHealthcare/scPROTEIN.git
cd scPROTEIN/
After downloading this repository, all the single-cell proteomics datasets used in our study will also be included. We provided these datasets in both .csv and .h5ad formats.
2.Setup of scPROTEIN python package
- We provided scPROTEIN package hosted on
PyPIand can be installed viapip.
pip install scprotein
- If for some reason this doesn't work on your device, you can also directly install scPROTEIN with the provided .whl file.
pip install docs/scprotein-0.1.1-py3-none-any.whl
- You can check if scPROTEIN package has been successfully installed via the following command:
python3 -c "import scprotein"
3.For datasets provided with raw peptide-level profile, scPROTEIN starts from stage 1 to learn the peptide uncertainty and obtain the protein-level abundance in an uncertainty-guided manner.
python3 train_stage1.py
After stage 1, the learned estimated peptide uncertainty array will be saved in folder './scPROTEIN'
4.Run stage 2 to obtain the learned cell embeddings.
python3 train_stage2.py --stage1 True
For datasets provided directly with the reconstructed protein-level profile, scPROTEIN will start from stage2.
python3 train_stage2.py
After stage 2, the learned cell embedding will be saved in folder './scPROTEIN/'.
For data integration analysis, you can firstly use function integrate_sc_proteomic_features to load datasets. Subsequently, the process of running scPROTEIN to learn cell embedding is similar. You can refer to the tutorials in data_integration for more details.
5.Evaluate the learned cell embeddings.
python3 visualization.py
After running visualization.py, a TSNE plot showing the cluster result will be saved in folder './scPROTEIN/', and a corresponding evaluation metric table will be displayed.
For loading checkpoints for scPROTEIN stage1 and stage2 on SCoPE2_Specht dataset for generating uncertainty and cell embedding, respectively:
python3 train_stage1.py --use_trained_scPROTEIN True
python3 train_stage2.py --stage1 True --use_trained_scPROTEIN True
The following notebooks are provided to show how to run scPROTEIN model
- tutorial_scPROTEIN_stage1 gives a detailed description for uncertainty estimation for scPROTEIN stage1.
- tutorial_scPROTEIN_stage2 provides an example using protein-level data from stage1 to learn cell embedding in stage2.
- data_integration shows the running process for data integration and batch correction across various MS acquisitions.
- downstream_application displays the analysis for clinical proteomic data, spatial proteomic data and cell cycle.
Hyperparameters for stage 1:
| Hyperparameter | Description | Default |
|---|---|---|
| batch_size | Batch_size | 256 |
| kernel_nums | Kernel num of each conv block | [300,200,100] |
| kernel_size | Kernel size of each conv block | [2,2,2] |
| max_pool_size | Max pooling size | 1 |
| conv_layers | Nums of conv layers | 3 |
| hidden_dim | Hidden dim for fc layer | 3000 |
Hyperparameters for stage 2:
| Hyperparameter | Description | Default |
|---|---|---|
| stage1 | If scPROTEIN starts from stage1 | False |
| num_hidden | Hidden dimension | 400 |
| num_proj_hidden | Dimension of projection head | 256 |
| num_layers | Number of GCN layers | 2 |
| num_protos | Number of prototypes | 2 |
| num_changed_edges | Number of added/removed edges | 10 |
| drop_edge_rate_1 | Dropedge rate for view1 | 0.2 |
| drop_edge_rate_2 | Dropedge rate for view2 | 0.4 |
| drop_feature_rate_1 | Mask_feature rate for view1 | 0.4 |
| drop_feature_rate_2 | Mask_feature rate for view1 | 0.2 |
| alpha | Balance factor | 0.05 |
| tau | Temperature coefficient | 0.4 |
Taking demo SCoPE2_Specht dataset (1490 cells, 3042 proteins) as an example, typical running time on a "normal" desktop computer is about 40 minutes for stage 1 and about 10 minutes for stage 2.
This tool is for research purpose and not approved for clinical use.
This is not an official Tencent product.
If you have any suggestions/ideas for scPROTEIN or have issues trying to use it, please don't hesitate to reach out to us. You can post an issue or reach us by email(nkuweili@mail.nankai.edu.cn, fionafyang@tencent.com).
Li, W., Yang, F., et al. A Versatile Deep Graph Contrastive Learning Framework for Single-cell Proteomics Embedding. https://www.biorxiv.org/content/10.1101/2022.12.14.520366v1
