GPLVM_Shaista

This is the repo for rotation in Teichmann Lab at Sanger Institute under the supervision of Emma Dann.

Conda environment setup

To create new conda environment with required packages for training the GPLVM model, run in terminal:

conda create -n gplvm_env python=3.7 
conda activate gplvm_env 
pip install torch==1.9.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html
pip install prettytable ipython numpy gpytorch==1.4.1 scanpy ipykernel  #gpytorch version 1.5.1?

To add the conda environment to the list of kernels available to jupyterLab, run:

python -m ipykernel install --user --name gplvm_env --display-name "Environment (gplvm_env)" 

Then reload the browser tab of JupyterLab to see the new environment in the list of available kernels.

Model document is at https://www.overleaf.com/project/605e07104b6c6221cfa7a557

Code Organisation:

The file TrainFunction.py contains two functions: run_model and run_model_randomInit to train an anndata object with 1 to 7 PC components or random values values. TrainFunction.py requires GPLVM model classes from file model.py.

model.py and demo_re_as_linear.py: copied from Aditya-edits branch of BGPLVM repo: https://github.com/vr308/BGPLVM_scRNA

run_model.py: python script to train GPLVM model for any anndata object:

usage: python run_model.py input initialisation output input: choose one of six dataset options (bonemarrow, gastrulation, forebrain, pancreas, pbmc, iPSC) or the path to your own anndata object. initialisation: choose one of two options: random or PCA output: name/path to output anndata file with trained GPLVM model