EFMSDTI

EFMSDTI integrates multi-source similarity networks and machine learning to predict drug-target interactions and improve DTI inference accuracy.

Key Features:

• Multi-Source Data Integration: Considers varying contributions of different data sources to optimize fusion for DTI prediction.
• Similarity Network Construction: Constructs 15 similarity networks derived from multi-source information and categorizes them into topological and semantic graphs for drugs and targets.
• Selective and Entropy Weighting with SNF: Applies Similarity Network Fusion (SNF) with selective and entropy weighting to integrate networks based on their contributions.
• Deep Learning Embedding: Uses a deep neural network to learn low-dimensional vector embeddings of drugs and targets.
• Predictive Modeling with LightGBM: Implements LightGBM (Gradient Boosting Decision Trees) for final DTI prediction.
• Performance Evaluation: Reported evaluation metrics include AUROC and AUPR of 0.982, and validation of 990 out of the first 1000 predicted DTIs.

Scientific Applications:

• Drug discovery and development: Prioritizes candidate drug-target pairs to support lead identification and optimization.
• Mechanism elucidation: Aids interpretation of potential drug mechanisms by predicting likely drug-target associations.
• Drug repurposing and target identification: Identifies potential new drugs and targets for disease treatment through predicted DTIs.

Methodology:

Build 15 similarity networks (topological and semantic) from multi-source data; apply Similarity Network Fusion (SNF) with selective and entropy weighting to integrate networks; train a deep neural network to learn low-dimensional drug and target embeddings; use LightGBM for final DTI prediction.