DeepsmirUD

DeepsmirUD predicts regulatory effects of small molecules on microRNA (miRNA) expression to support analysis of miRNA-mediated mechanisms and therapeutic targeting.

Key Features:

• Cross-Platform Prediction: Infers regulatory effects (upregulation or downregulation) of small molecules on miRNA expression across platforms.
• Deep Learning Frameworks: Leverages 12 state-of-the-art deep learning frameworks and reports AUC values of 0.843/0.984 and AUCPR values of 0.866/0.992 across two independent test datasets.
• Network Inference Approach: Employs a similarity-based network inference approach that achieved an accuracy of 0.813 for nearly 650 associated SM-miR relationships involving novel compounds and miRNAs.
• Integration with Cancer Data: Links miRNA-cancer relationship data connecting 1343 small molecules to 107 cancer diseases to support mechanism-of-action analysis and drug repositioning hypotheses.

Scientific Applications:

• Therapeutic Target Identification: Predicts small molecule effects on miRNA expression to aid identification of miRNA-related therapeutic targets.
• Drug Repositioning: Associates small molecules with cancer-related miRNAs to generate drug repositioning candidates for cancer indications.
• Mechanistic Insights: Provides predictions that inform mechanistic interpretation of how small molecules modulate miRNA-mediated regulatory pathways.

Methodology:

Combines 12 deep learning frameworks with a similarity-based network inference approach to predict impacts of small molecules on miRNA expression.