monae: multi-modal single-cell integration and imputation

Environment

Python version: python 3.8

Install the following dependencies via pip or conda:

pip install muon
pip install torchbiggraph
pip install scanpy
pip install pytorch-ignite
pip install typing_extensions
pip install tensorboardX
pip install torch-1.11.0+cu113

Dataset

Dataset processed (dataset_monae)：

link: https://pan.baidu.com/s/1lOjSVVIUQ3WmqTDUJzdxMQ?pwd=1etw

For the raw data, select annotation file in preprocess_1.py:

gtf="./src/bedtools/annotation/gencode.v42.chr_patch_hapl_scaff.annotation.gtf.gz", # Human
gtf="./src/bedtools/annotation/gencode.vM25.chr_patch_hapl_scaff.annotation.gtf.gz", # Mouse

Example

Trained Monae

See the monae example in the notebook_monae. Trained Monae model:

link: https://pan.baidu.com/s/1F-EzYazz31_LEOMEfxMM2w?pwd=v8wl

The models and datasets are organized as follows:

--src
...
--notebook_monae
  --dataset_monae
  --model # trained monae model
  --integration.ipynb # for integration
  --imputation_rna2atac.ipynb # for cross-modal imputation (rna->atac)
  --imputation_atac2rna.ipynb # for cross-modal imputation (atac->rna)

In ipynb files, select one of four datasets: pbmc, chen2019, ma2020, muto2021.

Auto infer guidance

A new version is provided here that can generate guidance without relying on annotation files, please refer to the dir liu. Run the following files in sequence:

1. 1_train_gnn.py
2. 2_graph_generation.py
3. 3_integration.py
4. 4_inference.ipynb

Other modality

Same order of execution as above, here are examples for RNA and protein modalities, please refer to the dir prot.

Monae-Extension

At present, the repo released a stable and convenient training strategy for Monae's variant Monae-E. Run the following files in sequence:

1. preprocess_1.py: data preprocess and constructing graph guidance.
2. preprocess_2.py: Unified learning of cell and feature representations.
3. integration.py: Train model.

Take the PBMC dataset as an example, the trained model has been saved here: dataset/pbmc/result/ckpt/monae.dill. Loading trained model in inference.ipynb.

Todo

• Add new preprocessing method: rank value encoding, like Geneformer.
• Complete training process for Monae and Monae-E
• Pipeline for generating gene networks

Reference

[1] Cellcano: supervised cell type identification for single cell ATAC-seq data, nature communications, 2023

https://github.com/marvinquiet/Cellcano

[2] Multi-omics single-cell data integration and regulatory inference with graph-linked embedding, nature biotechnology, 2022

https://github.com/gao-lab/GLUE

[3] SIMBA: single-cell embedding along with features, nature methods, 2023

https://github.com/pinellolab/simba

[4] scGPT: toward building a foundation model for single-cell multi-omics using generative AI, nature methods, 2024

https://github.com/bowang-lab/scGPT