LTGN

Loss Trim Graph-based Network (LTGN), an unsupervised framework based on the interpretability of graph neural networks.

Installation

The LTGN package is developed based on the Python libraries PyTorch and PyG (PyTorch Geometric) framework, and can be run on GPU (recommend) or CPU. Before installing stVGP, please ensure that PyTorch, and PyG (PyTorch Geometric) are already installed. These dependencies are required for LTGN to function properly, but they are not automatically installed during the installation process to allow greater flexibility.

# Installation
pip install LTGN

# Requirement
# scipy
# numpy
# pandas
# scikit-learn
# pingouin

Quick-start tutorial

Here, we provide guidance on using the LTGN sample data to help you quickly get started with our method. Here we use Beeline data as an example. The data can be found in the data folder (https://github.com/wzdrgi/LTGN/tree/main/data/Beeline/mDC), and the PPI information can be obtained from the link https://figshare.com/articles/dataset/LTGN_PPI/31570009.

Import packages

import pandas as pd
import numpy as np
import torch
import os
from sklearn.preprocessing import RobustScaler
import LTGN

Set parameters

seed = 0
top = 500
threshold = 0
epoch = 550
std_scaler = RobustScaler()
lr = 0.05
ln = 5
m = 'ppi'
ppthreshold = 0
ref = None
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')

data and processing

datafile = 'F:\GitDemo\LTGN\data\Beeline/mDC/ExpressionData.csv'
celltype_data = pd.read_csv(datafile,index_col=0).T
order_name  = 'F:/GitDemo/LTGN/data/Beeline/mDC/GeneOrdering.csv'
order = pd.read_csv(order_name,index_col=0)
sorted_order = order.sort_values(by='Variance', ascending=False)
topgene = list(sorted_order.iloc[:top,].index)
selected = celltype_data.loc[:,topgene]
numeric_cols = selected.select_dtypes(include=['number']).columns
selected[numeric_cols] = std_scaler.fit_transform(selected[numeric_cols])
filename = 'mDC' + str(threshold)+ 'seed' + str(seed) + str(epoch) +' ln' + str(ln) + 'top' + str(top) + 'ppth' + str(ppthreshold)+ m + '.csv'

Run

edgee_df = LTGN.oneFeaturemodel_ppi(
                        data_df = selected,
                        lr = 0.05,
                        epoch = 1,
                        # Please modify the training epoch.
                        threshold = threshold,
                        ppthreshold = ppthreshold,
                        seed = seed,
                        ln = ln,
                        ref = ref,
                        running_name = 'mDC' + '_rb' + '_top' + str(top) + '_ppth' + str(ppthreshold) + '_seed' + str(seed),
                        m = m,
                        batch_size = None,
                        device = device,
                        spe = 'homo')