We trained our sub-models on the dataset of Drugvirus info 2.0 .
Advanced Features Extractors
Pre-trained deep learning models were leveraged as feature extractors.
Multi-Modal Deep Embeddings
We leveraged pre-trained deep learning models such as Resnet50, Resnet101, EfficientNet, and Inception_Resne as image-based feature extractors, Albert, Roberta, Bert, and Gpt-2 as corpus-based transformers. We also leveraged sequence-based feature extractors such as ilearn and doc2vec and network-based feature extractors such as role2vec and node2vec. Those multi-model embeddings from multi-feature view domains can help the classifier to conduct the prediction from a higher point of view.
Expandable
Our model is an expandable platform. On the one hand, our model can predict novel viruses with fasta files or descriptions. On the other hand, our methods can expand its drug and virus candidates set easily, which means as more pairs are collected, our model can also upgrade simultaneously.
Interface
An interface sotaware is available to predict drug candidates from Viral DNA Fasta files/Descriptions.