WTHMDA 0.0.4

WTHMDA

WTHMDA

Contents

• Introduction
• Package requirement
• Installation
• Data preprocess
• Model training and prediction
• Citation
• Contact

Introduction

In this study, we propose a novel deep learning framework named WTHMDA (Weighted Taxonomic Heterogeneous-network based Microbe-Disease Association), which leverages a weighted graph convolution network and microbial taxonomy common tree for microbe-disease association prediction.

Package requirement

• torch==1.9.0
• dgl==0.9.1
• ete3==3.1.3
• gensim==4.0.1

Installation

setup.py

Then all tools are located at ‘bin’ folder:

get_disease_feature.py // To generate disease feature utilizing weighted Deepwalk

get_microbe_feature.py // To generate microbe feature utilizing weightedDeepwalk

get_edges.py // To generate directed microbe links and four types of weighted microbe ontological similarities

WTHMDA_run.py // For model training and prediction

Data preprocess

To generate disease feature utilizing weighted Deepwalk, one file are required as input including integrated disease functional similarities.
a. integrated disease functional similarities (required)

1	0.21	0.52	0.11
0.21	1	0	0.44
0.52	0	1	0.87
0.11	0.44	0.87	1

You can specify the integrated disease functional similarities path (D_feat_path) as input.

from WTHMDA2.example.get_disease_feature import get_disease_feature

get_disease_feature(D_feat_path)

To generate microbe feature utilizing weighted Deepwalk, one file are required as input including integrated microbe functional similarities. a. integrated microbe functional similarities (required)

1	0.75	0.9	0.43
0.75	1	0.71	0.1
0.9	0.71	1	0.53
0.43	0.1	0.53	1

You can specify the integrated microbe functional similarities path (M_feat_path) as input.

from WTHMDA2.example.get_microbe_feature import get_microbe_feature

get_microbe_feature(M_feat_path)

To generate microbe feature utilizing weighted Deepwalk, one file are required as input including integrated microbe functional similarities. a. integrated microbe functional similarities (required)

0	Abiotrophia
1	Bacteroides
2	Caloramator
3	Dermacoccus

You can specify the microbe list path (MicroPd_path) as input.

from WTHMDA2.example.get_edges import get_edges

get_edges(MicroPd_path)

Model training and prediction

To train WTHMDA model and prediction, two files are required as input including train_edge and test_edge. a. train_edge (required)

disease	microbe	label
1	108	1
4	43	0
36	230	1

b. test_edge (required)

disease	microbe	label
3	165	1
15	62	1
14	42	0

Then, you can predict the status of microbiomes using the model generated by the training procedure.

from WTHMDA2.example.WTHMDA_example import WTHMDA_example

WTHMDA_example(train_edge_path = 'data/example/train_edge.csv',
                test_edge_path = 'data/example/test_edge.csv',
                save_path = 'result/example')

For convenience,We set an example dataset(example) in ‘data/’ folder.

Please create your code file in the same directory as the "data" folder, like the following example:

├── data
│   └── example
└── your_code.py

You can run the following code to get a quick start.

from WTHMDA2.example.WTHMDA_example import WTHMDA_example
from WTHMDA2.example.get_microbe_feature import get_microbe_feature
from WTHMDA2.example.get_edges import get_edges
from WTHMDA2.example.get_disease_feature import get_disease_feature

get_disease_feature(D_feat_path = 'data/example/disease_I_S.csv')
get_edges(MicroPd_path = 'data/example/microbe_list.xlsx')
get_microbe_feature(M_feat_path = 'data/example/microbe_I_S.csv')
WTHMDA_example(train_edge_path = 'data/example/train_edge.csv',
                test_edge_path = 'data/example/test_edge.csv',
                save_path = 'result/example'

Citation

Contact

All problems please contact WTHMDA development team: Xiaoquan Su    Email: [e-mail]