Due to GitHub's file size limitations, the trained models for EVLncRNA-net can be accessed at the following link:

Google Drive: EVLncRNA-net Trained Models

Please download the models from the provided Google Drive folder and place the model folder into your code directory.

Bio==1.5.8
matplotlib==3.1.2
numpy==1.23.5
pandas==1.5.3
scikit_learn==1.2.2
torch==2.0.0
torchvision==0.15.0
tqdm==4.65.0
torch==2.1.2+cu121
torch-cluster==1.6.3+pt21cu121
torch_geometric==2.5.3

In config.py, change the dataset by modifying the self.data_name parameter:

self.data_name = 'plant'

You only need to adjust this parameter, with options including human, mouse, and plant.

Specify the paths for the training and test sets in config.py.

Run train.py:

data = Biodata(fasta_file=config.train_x, label_file=config.train_label,
               feature_file=config.train_fea)
dataset = data.encode(thread=20)

This handles dataset processing.

if config.data_name == 'human':
    model = mynet(label_num=2, K=3, d=3, node_hidden_dim=3, other_feature_dim=128, other_feature_dim_in=2000, pnode_nn=False, fnode_nn=False).to(device)  # human_net
else:
    model = mynet(label_num=2, K=3, d=3, node_hidden_dim=3, other_feature_dim=128, other_feature_dim_in=4000, data_name=config.data_name).to(device)  # mouse & plant net

This sets up the model architecture.

train(dataset, model, weighted_sampling=False, batch_size=config.batch_size,
      learning_rate=config.learning_rate, epoch_n=20, iffloss=1)

This initiates the training process.

Where:

• iffloss=1 uses FocalLoss.
• Otherwise, CrossEntropyLoss is used as the loss function.

After downloading the pretrained models from the model folder, select the desired species model in config.py and run Test.py to generate predictions.

Let me know if you need further adjustments!