PNETwa-MLP

Table of Contents

• Introduction
• Install model
  • Step 1
  • Step 2
• Training model
  • Step 1
  • Step 2
  • Step 3
  • Step 4
  • Step 5
  • Step 6
  • Step 7

Introduction

In this study, we propose a novel approach that combines the incorporation of biological information with a Graph Neural Network called P-NET and an MLP. we integrated multiple datasets from Chinese and Western prostate cancer patients, a total of 1360 whole exome/genome alteration data for training and testing, and a set of 115 patients' data as an external validation dataset. We used mixed Chinese and Western data to train a P-NET (hereinafter referred to as PNETwa).PNETwa is a type of Graph Neural Network that possesses the ability to learn node-level and graph-level representations while considering the topological relationships between nodes. The MLP, on the other hand, is a commonly used feedforward neural network model known for its powerful nonlinear modeling capabilities. Together, they can learn more comprehensive representations of global features from the node-level or graph-level representations extracted by the GNN.

We used PNETwa to extract 18 signature genes associated with prostate cancer metastasis from genome-wide alteration data. These signature genes are then input into the MLP to realize the prediction of prostate cancer metastasis. Our method achieves an accuracy of 0.87 and an F1 score of 0.84 in data from Western populations, and an accuracy of 0.88 and an F1 score of 0.75 in mixed data from Western and Asian populations.

Overall, this research contributes to the effective resolution of complex issues in primary and metastatic detection in clinical settings, opening up new possibilities in this field.

Install model

Step 1

Create a PNETwa environment：(ubuntu)

cd ./PNETwa
conda env create --name PNETwa --file=environment.yml

Step 2

Create a MLP environment：(windows)

cd ./MLP
conda env create --name MLP --file=environment.yml

Training model

Step 1

Activate the PNETwa environment:

cd ./PNETwa
source activate PNETwa

Step 2

Train PNETwa:

export PYTHONPATH=~/pnet_prostate_paper:$PYTHONPATH
cd ./PNETwa/train
python run_me.py

Step 3

Get node contribution:

cd ./PNETwa/analysis
python run_it_all.py

This will generate 'node_importance_graph_adjusted.csv' in './PNETwa/analysis/extracted/'

Step 4

To extract the contribution magnitude of genes (coef_combined column) from 'node_importance_graph_adjusted.csv', follow these steps:

1. Open the file 'node_importance_graph_adjusted.csv'
2. Locate the column named "coef_combined," which represents the contribution magnitude of genes.
3. Retrieve the values from the coef_combined column, which indicate the respective contribution magnitudes of genes.

Here is an example Python code snippet that demonstrates how to extract the values from the "coef_combined" column using the pandas library:

# Import the pandas library
import pandas as pd

# Load the CSV file into a pandas DataFrame
df = pd.read_csv("node_importance_graph_adjusted.csv")

# Filter rows where the 'layer' column has a value of 1
filtered_df = df[df["layer"] == 1]

# Sort the DataFrame based on the 'coef_combined' column in descending order
sorted_df = filtered_df.sort_values("coef_combined", ascending=False)

# Print the sorted DataFrame
print(sorted_df)

Step 5

Activate the MLP environment:

conda activate MLP

Step 6

Train MLP:

cd ./MLP
python model.py

To modify the feature genes in the 'ls' function in the model.py file, you can follow these steps:

1. Open the model.py file in a text editor or integrated development environment (IDE).
2. Locate the 'ls' function within the file.
3. Identify the feature genes that you want to modify. These may be variables, parameters, or other elements that affect the behavior of the function.
4. Save the model.py.

Step 7

Local deployment of webui

cd ./web
python web.py