BioChemDDI

BioChemDDI predicts drug-drug interactions (DDIs) by integrating chemical sequence information and biological function similarity to identify potential interaction patterns relevant to drug safety and therapeutic co-administration.

Key Features:

• Integration of Multi-Level Information: Integrates diverse biological and chemical information and captures chemical sequence data using Natural Language Processing (NLP).
• Similarity Network Fusion (SNF): Applies Similarity Network Fusion (SNF) to fuse multiple biological function similarity datasets.
• Hierarchical Representation Learning (HARP): Leverages Hierarchical Representation Learning for Networks (HARP) to extract deep network structural information.
• Self-Attention Feature Descriptor: Employs a self-attention module to construct comprehensive feature descriptors that integrate biochemical and network features.
• Deep Neural Network (DNN) Prediction: Uses a deep neural network (DNN) to generate interaction predictions and reported performance that outperforms previous models.
• Graph Collapse Technique: Introduces graph collapse during network structure extraction to capture complex interaction networks.
• Biochemical Pre-Training: Utilizes biochemical information in the pre-training process to enhance predictive power.

Scientific Applications:

• Breast Cancer: 24 of the top 30 predicted drugs related to breast cancer were confirmed against existing databases.
• Hepatocellular Carcinoma: 18 of the top 30 predicted drugs were verified.
• Malignancies: 20 of the top 30 predicted drugs were validated.

Methodology:

Integration of Natural Language Processing (NLP), Similarity Network Fusion (SNF), Hierarchical Representation Learning for Networks (HARP), self-attention mechanisms, graph collapse during network extraction, biochemical pre-training, and a deep neural network (DNN) for prediction.