Our Software and test data have been archived at the dedicated repository of Zenodo. The archived Software version is available here
GCATSL is a deep learning model that can be used for SL prediction. As shown in the following flowchart (a), GCATSL first learns representations for nodes based on different feature graphs together with a known SL interaction graph, and then uses the learned node representations to reconstruct SL interaction matrix for SL prediction.
GCATSL is implemented with Tensorflow library. For the detail instruction of installing Tensorflow, see the guidence on official website of Tensorflow.
You'll need to install the following packages in order to run the codes.
- Python 3.7
- Tensorflow 1.13.1
- numpy 1.16.2
- scipy 1.4.1
- sklearn
This Python script is designed to implement the GCATSL model. GCATSL model needs two kinds of data files in TXT format as inputs, i.e., adj.txt and feature_x.txt. adj.txt represents the file of known SL pairs, where the first and second columns denote the numbers of gene1 and gene2 (e.g., 1,2,3...) respectively, and the third column deontes the labels of SL pairs (i.e., 1 or 0). A toy example for adj.txt is available below, where the second row means that the first gene is confirmed to be associated with the fifth gene.
A toy example of adj.txt
| gene1 | gene2 | label |
|---|---|---|
| 1 | 5 | 1 |
| 3 | 9 | 1 |
| 5 | 30 | 1 |
Note that for convenient instruction, here we add names of genes and their label in the first line in the above example. You should remove these information when inputting them to the model.
GCATSL can use multiple feature graphs of genes for SL prediction. feature_x.txt is a matrix that represents the 'x'-th feature graph for genes. It should be noted that the dimensions of all feature graphs should be completely consistent. To reproduce the results for SL prediction, here we provided a set of input data (http://synlethdb.sist.shanghaitech.edu.cn/downloadPage.php) as examples, which are available in ./data/toy_examples. Please see the readme for the detailed explanations about the example data.
The outputs of GCATSL model include two files, i.e., test_result.txt and log.txt. test_result.txt records the predicted scores for test samples. The first, second, third and fourth columns in the test_result.txt denote the number of gene1, the number of gene2, the predicted score for gene1-gene2 SL pair and their label, respectively. A toy example of test_result.txt is available below.
A toy example of test_result.txt
| gene1 | gene2 | score | label |
|---|---|---|---|
| 1 | 5 | 0.924 | 1 |
| 3 | 9 | 0.865 | 1 |
| 5 | 30 | 0.532 | 1 |
In addition, we used metrics AUC and AUPR to evaluate our proposed GCATSL model. log.txt records the values of metrics AUC and AUPR for each fold CV (Cross Validation) respectively.
First, you need to clone the repository or download source codes and data files.
$ git clone https://github.com/longyahui/GCATSL.git
You can see the optional arguments for GCATSL by "help" option:
$ python source/main.py --help
You can easily run GCATSL with two steps, and then obtain the predicted results.
-
Unzip the file
./data/toy_examples/global interaction matrix.rarin directory./data/toy_examples/. -
Run
main.pyto obtain the predicted results as follows:python source/main.py --n_epoch 600 \ --n_head 2 \ --n_fold 5 \ --n_node 6375 \ --n_feature 3 \ --learning_rate 0.005 \ --weight_decay 0.0001 \ --dropout 0.7 \ --input_dir ../data/toy_examples/ \ --output_dir ../output/ \ --log_dir ../output/ \
yahuilong@hnu.edu.cn