Experimental setup

Follow these steps to setup for reproducing the experiments provided in ....

1. Install Singularity from https://docs.sylabs.io/guides/3.0/user-guide/installation.html:

   • Install Singularity release 3.10.2, with Go version 1.18.4
   • Suggestion: follow instructions provided in Download and install singularity from a release section after installing Go
   • Install dependencies from: https://docs.sylabs.io/guides/main/admin-guide/installation.html

2. Clone the lstmeta-tnf repository in your home folder

git clone https://github.com/smilies-polito/lstmeta-tnf.git #change the link

3. Move to the lstmeta-tnf source subfolder, and build the Singularity container with

cd  lstmeta-tnf/source
sudo singularity build lstmeta-tnf.sif lstmeta-tnf.def

or using fake root privileges

cd lstmeta-tnf/source
singularity build --fakeroot lstmeta-tnf.sif lstmeta-tnf.def

Reproducing the analysis interactively within the lstmeta-tnf Singularity container

To run analyses manually launch the lstmeta-tnf Singularity container. Move to the source folder, and launch the scripts as follows.

First of all, launch the lstmeta-tnf Singularity container

cd source
singularity shell --fakeroot --nv --bind /path/to/your/data/folder:/mnt lstmeta-tnf.sif

This will run a shell within the container, and the following prompt should appear:

Singularity>

Then open a bash shell by running

bash

Now follow the steps below.

Dataset creation

Move to the simulations folder and execute the data_extraction_definitive.py script

cd lstmeta-tnf/simulations
python data_extraction_definitive.py

the data_extraction_definitive.py execute the 7,288 different simulations and saves the input parameters and the cell states of each simulation respectively in the /path/to/your/data/folder/data/input_parameters and /path/to/your/data/folder/data/cell_data folders.

Dataset analysis and CSVs creation

Move to the metamodel folder and run the data_exploration.ipynb notebook. From the terminal

cd ../metamodel
jupyter nbconvert --execute --to notebook --inplace data_exploration.ipynb

Metamodel training

You can train a new metamodel LSTM by running the rec_model_train.py script. You can also specify the tumor_radius (i.e. 50, 100, 275, 400), the highest latent dimension latent_dimension (default 1000) of the model, the window length window_size (default 24) of the sequences, the number of latent layers n_layers_in (default 2) in the encoder and decoder, the linear dropout dropout_lin (default 0), and the recurrent dropout dropout_rec (default 0.2).

python rec_model_train.py --tumor_radius 50 --latent_dim 1000 --window_size 24 --n_layers_in 2 --dropout_lin 0 --dropout_rec 0.2

The results will be saved in a new metamodel/experiments folder.

Reproducing the analysis running the Singularity container

To reproduce the analysis from this paper, run the singularity container lstmeta-tnf.sif

Move to the source folder and run the lstmeta-tnf.sif file

cd source
singularity run --fakeroot --nv --bind /path/to/your/data/folder:/mnt lstmeta-tnf.sif