DeepdlncUD

DeepdlncUD predicts whether small molecules upregulate or downregulate long non-coding RNA (lncRNA) expression to support identification and prioritization of lncRNA-targeting therapeutic candidates.

Key Features:

• Integration of Multiple Algorithms: Uses nine deep learning algorithms to infer regulation types for SM-lncR pairs.
• Extensive Feature Encoding: Encodes 1369 features per SM-lncR pair, including sequence data, representational information, and physiochemical properties.
• Performance Metrics: Optimized on a training dataset of 771 upregulation and 739 downregulation pairs and reports AUC 0.674, AUCPR 0.722, F1-score 0.681, and Jaccard Index 0.516 on a test set of 222 SM-lncR pairs.
• Pathological Context Accuracy: Achieves an accuracy of 0.700 for identifying small molecules that regulate disease-associated lncRNAs.
• Connectivity Scoring for Drug Prioritization: Computes connectivity scores to estimate small-molecule potential as drugs targeting lncRNA-regulated diseases, with approximately half of predictions aligning with known drug targets.

Scientific Applications:

• Regulation Prediction: Predicts upregulation or downregulation of lncRNAs by small molecules for hypothesis generation.
• SM-lncR Interaction Screening: Prioritizes SM-lncR pairs for experimental validation in drug-discovery workflows.
• Disease-focused Discovery: Identifies small molecules that modulate disease-associated lncRNAs to support therapeutic candidate selection.
• Mechanistic Support: Provides computational evidence of regulatory relationships between small molecules and lncRNAs to inform targeted therapy development.

Methodology:

Encodes 1369 features per SM-lncR pair, applies nine deep learning algorithms with systematic optimization on a training set of 771 upregulation and 739 downregulation pairs, evaluates on a test set of 222 SM-lncR pairs, and computes connectivity scores for drug prioritization.