#GCNLA

---

GCNLA overview

GCNLA is used to infer cell-cell interactions based on transcriptomics data and spatial location information. In this work, we propose a network architecture based on graph convolution network and long short-term memory attention module-GCNLA, which contains a graph convolution layer, a long short-term memory network, an attention module, and residual connections.

Installation & Setup

Make sure to clone this repository along with the submodules as follows:

git clone --recurse-submodules https://github.com/sharonycc/GCNLA
cd GCNLA

To install dependencies to a conda environment, follow the instructions provided: First, create a basic conda environment with python 3.8.18

conda create --name GCNLA python=3.8.18

Now, activate your environment and utilize the requirements.txt file to install non pytorch dependencies

conda activate GCNLA
pip install -r requirements.txt

Data

Three datasets were utilized for evaluation:

1. seqFISH profile of mouse visual cortex (Zhu et al., 2018)
2. MERFISH profile of mouse hypothalamic preoptic region (Moffitt et al., 2018)

All of the preprocessed data are organized into pandas dataframes and are located at ./data. These dataframes can be used directly as input to GCNLA.

Run GCNLA

To run Clarify, run main.py and configure parameters based on their definitions below:

usage: main.py [-h] [-m MODE] [-i INPUTDIRPATH] [-o OUTPUTDIRPATH] [-s STUDYNAME] [-t SPLIT] 
               [-n NUMGENESPERCELL] [-k NEARESTNEIGHBORS] [-l LRDATABASE] [--fp FP] [--fn FN] [-a OWNADJACENCYPATH]

The first row of parameters are necessary

• -m MODE, --mode MODE clarify mode: preprocess,train (pick one or both separated by a comma)
• -i INPUTDIRPATH, --inputdirpath Input directory path where ST dataframe is stored
• -o OUTPUTDIRPATH, --outputdirpath Output directory path where results will be stored
• -s STUDYNAME, --studyname clarify study name to act as identifier for outputs
• -t SPLIT, --split ratio of test edges [0,1)

This second row of parameters have defaults set and are not needed.

• -n NUMGENESPERCELL, --numgenespercell Number of genes in each gene regulatory network (default 45)
• -k NEARESTNEIGHBORS, --nearestneighbors Number of nearest neighbors for each cell (default 5)
• -l LRDATABASE, --lrdatabase 0/1/2 for which Ligand-Receptor Database to use (default 0 corresponds to mouse DB)
• --fp FP (experimentation only) add # of fake edges to train set [0,1)
• --fn FN (experimentation only) remove # of real edges from train set [0,1)
• -a OWNADJACENCYPATH, --ownadjacencypath Using your own cell level adjacency (give path)

For example, if you wanted to run Clarify (both preprocessing and training) on the seqFISH data input with a 70/30 train-test split, then use the following command and set the output folder and studyname accordingly:

python main.py -m preprocess,train -i ./data/MERFISH/merfish_dataframe.csv -o [OUTPUT FOLDER PATH] -s [STUDYNAME] -t 0.3