This package is used for protein-protein interaction (PPI) prediction
・LSTM-PHV predicts PPI by amino acid sequences alone.
・Attention weights could be useful to find binding interfaces.
Python : 3.8.0
Anaconda : 4.9.2
※We recommend creating virtual environments by using anaconda.
This CLI system is used for three processing as follows.
・Training of a word2vec embedding model to encode amino acid sequences.
・Training of LSTM-PHV model for PPI prediction.
・PPI prediction.
0-1. Creating a virtual environment.
$conda create -n [virtual environment name] python==3.8.0
ex)
$conda create -n lstm_phv_network python==3.8.0
0-2. Activating the virtual environment
$ conda activate [virtual environment name]
ex)
$ conda activate lstm_phv_network
Execute the following command in the directory where the package is located.
$pip install ./LSTM-PHV/dist/LSTM-PHV-0.0.1.tar.gz
A word2vec model can be trained by following command.
$lstmphv train_w2v -i [Training data file path (fasta format)] -o [output dir path]
ex)
$lstmphv train_w2v -i ~/LSTM-PHV/sample_data/w2v_sample_data.fa -o ~/LSTM-PHV/w2v_model
other options)
| option | explanation | necessary or not | default value |
|---|---|---|---|
| -i (--import_file) | Path of training data (.fasta) | necessary | - |
| -o (--out_dir) | Directory to save w2v model | necessary | - |
| -k_mer (--k_mer) | Size of k in k_mer | not necessary | 4 |
| -v_s (--vector_size) | Vector size | not necessary | 128 |
| -w_s (--window_size) | Window size | not necessary | 3 |
| -iter (--iteration) | Iteration of training | not necessary | 1000 |
(Results)
Model files will be output to the specified directory.
Filename: AA_model.pt, AA_model.pt.trainables.syn1neg.npy, AA_model.pt.wv.vectors.npy
| Filename | contents |
|---|---|
| AA_model.pt | word2vec model file |
| AA_model.pt.trainables.syn1neg.npy | word2vec model file (depending on the model size) |
| AA_model.pt.wv.vectors.npy | word2vec model file (depending on the model size) |
LSTM-PHV model for PPI prediction can be trained by following command (Promote the use of GPU-enabled environments).
$lstmphv train_deep -t [Training data file path (csv format)] -v [Training data file path (csv format)] -w [word2vec model file path] -o [output dir path]
ex)
$lstmphv train_deep -t ~/LSTM-PHV/sample_data/sample_training_data.csv -v ~/LSTM-PHV/sample_data/sample_validation_data.csv -w ~/LSTM-PHV/w2v_model/AA_model.pt -o ~/LSTM-PHV/deep_model
Note that csv files need to contein the following contents (Check the sample data at /LSTM-PHV/sample_data)
First column:human protein IDs
Second column:viral protein IDs
Third column:human protein sequences
Forth column:viral protein sequences
Fifth column:label (1: interact, 0: not interact)
other options)
| option | explanation | necessary or not | default value |
|---|---|---|---|
| -t (--training_file) | Path of training data file (.csv) | necessary | - |
| -v (--validation_file) | Path of validation data file (.csv) | necessary | - |
| -w (--w2v_model_file) | Path of a trained word2vec model | necessary | - |
| -o (--out_dir) | Directory to output results | necessary | - |
| -l (--losstype) | Loss type (imbalanced: loss function for imbalanced data, balanced: Loss function for balanced data) | not necessary | imbalanced |
| -t_batch (--training_batch_size) | Training batch size | not necessary | 1024 |
| -v_batch (--validation_batch_size) | Validation batch size | not necessary | 1024 |
| -lr (--learning_rate) | Learning rate | not necessary | 0.001 |
| -max_epoch (--max_epoch_num) | Maximum epoch number | not necessary | 10000 |
| -stop_epoch (--early_stopping_epoch_num) | Epoch number for early stopping | not necessary | 20 |
| -thr (--threshold) | Threshold to determined whether interact or not | not necessary | 0.5 |
| -k_mer (--k_mer) | Size of k in k_mer | not necessary | 4 |
(Results)
Text and model files will be output to the specified directory.
Filename: model/deep_model and deep_HV_result.txt
| Filename | contents |
|---|---|
| model/deep_model | LSTM-PHV model file |
| deep_HV_result.txt | Documenting the learning process |
PPI prediction is executed by following command (Promote the use of GPU-enabled environments).
$lstmphv predict -i [data file path (csv format)] -o [output dir path] -w [word2vec model file path] -d [deep learning model file path]
ex)
$lstmphv predict -i ~/LSTM-PHV/sample_data/sample_test_data.csv -o ~/LSTM-PHV/results -w ~/LSTM-PHV/w2v_model/AA_model.pt -d ~/LSTM-PHV/deep_model/deep_model
other options)
| option | explanation | necessary or not | default value |
|---|---|---|---|
| -i (--import_file) | Path of data file (.csv) | necessary | - |
| -o (--out_dir) | Directory to output results | necessary | - |
| -w (--w2v_model_file) | Path of a trained word2vec model | necessary | - |
| -d (--deep_model_file) | Path of a trained lstm-phv model | necessary | - |
| -thr (--threshold) | Threshold to determined whether interact or not | not necessary | 0.5 |
| -batch (--batch_size) | Batch size | not necessary | 1024 |
| -k_mer (--k_mer) | Size of k in k_mer | not necessary | 4 |
Note that csv files need to contein the following contents (Check the sample data at /LSTM-PHV/sample_data)
First column:human protein IDs
Second column:viral protein IDs
Third column:human protein sequences
Forth column:viral protein sequences
(Results)
CSV files will be output to the specified directory.
Filename: result.csv, human_transformed_vec.csv, viral_transformed_vec.csv, human_protein_attention_weights.csv, viral_protein_attention_weights.csv
| Filename | contents |
|---|---|
| result.csv | Predictive scores and labels |
| human_transformed_vec.csv | Transformed vector generated by the LSTM-PHV network while extracting human protein features. Each row contains the transformed vector in a sample |
| viral_transformed_vec.csv | Transformed vector generated by the LSTM-PHV network while extracting viral protein features. Each row contains the transformed vector in a sample |
| human_protein_attention_weights.csv | Attention weights generated by the LSTM-PHV network while extracting human protein features. Each row contains the attention weights in a sample |
| viral_protein_attention_weights.csv | Attention weights generated by the LSTM-PHV network while extracting viral protein features. Each row contains the attention weights in a sample |
We provided sample data, word2vec model, and LSTM-PHV model as well as CLI system. Note that sample data is not the benchmark datasets and this is only present the example.