DeepR2cov

DeepR2cov identifies potential anti-inflammatory therapeutic agents for COVID-19 by applying a deep representation model to heterogeneous drug networks integrated from eight distinct datasets and linking network-derived embeddings with transcriptomics.

Key Features:

• Heterogeneous Drug Network Integration: Constructs an integrated heterogeneous drug network by combining eight distinct drug-related datasets.
• Multi-Hub Characteristic Exploration: Employs 3 billion special meta paths to capture long-range structural dependencies and intricate semantic relationships in multi-hub networks.
• Deep Representation Model: Learns low-dimensional node embeddings that encode both structural and semantic information from the integrated network.
• Predictive Analysis: Combines learned representation vectors with transcriptomics data to predict 22 candidate drugs targeting inflammatory markers such as tumor necrosis factor-α and interleukin-6.
• Validation of Predictions: Validates predicted therapeutic associations using PubMed literature, ongoing clinical trials, and molecular docking simulations.
• Performance Evaluation: Evaluated across five biomedical applications and compared to five existing network representation approaches, demonstrating superior performance.

Scientific Applications:

• Identification of anti-inflammatory therapeutics for COVID-19: Prioritizes drugs that may modulate hyperinflammation in COVID-19 by targeting pathways including tumor necrosis factor-α and interleukin-6.
• Drug repurposing and discovery: Nominates candidate drugs for experimental follow-up by integrating network representations with transcriptomic signatures.
• Systems-level analysis of drug–disease interactions: Enables study of complex, multi-hub biological interactions relevant to inflammatory responses using integrated heterogeneous networks.

Methodology:

Integrates eight drug-related datasets into a heterogeneous network, generates and leverages 3 billion special meta paths to train a deep representation model that produces low-dimensional embeddings, combines these embeddings with transcriptomics for drug prediction, and validates associations using PubMed, clinical trial data, and molecular docking while benchmarking across five biomedical applications against five existing network representation methods.