SG-LSTM

SG-LSTM predicts miRNA-gene associations by combining sequence and geometrical features with Long Short-Term Memory (LSTM) networks to identify regulatory interactions relevant to gene expression and disease.

Key Features:

• Integration of Sequence and Geometrical Information: Generates representational features for miRNAs and genes by extracting sequence-based and geometrical information to capture complex interaction determinants.
• Deep Learning Approach (LSTM): Employs Long Short-Term Memory networks to model sequential and contextual patterns in the generated features for association prediction.
• Superior Performance: Demonstrated an area under the curve (AUC) of 0.94 in comparative evaluations against state-of-the-art methods.
• Scalability and Large Dataset Handling: Validated on a dataset comprising 14,668,698 miRNA-gene pairs, indicating capacity for large-scale analysis.
• Case Studies and Recommendations: Produces prioritized candidate genes for specific miRNAs, exemplified by top-10 recommendations for hsa-miR-335-5p.

Scientific Applications:

• Cancer Research: Facilitates identification of miRNA-mediated regulatory interactions implicated in cancer development and progression.
• miRNA Regulatory Mechanism Studies: Supports investigation of miRNA-gene regulatory networks to inform therapeutic target prioritization and personalized medicine approaches.

Methodology:

Feature generation: extracting sequence-based and geometrical features for miRNAs and genes; model training: employing LSTM networks to learn from generated features and predict associations; validation and testing: evaluating performance using metrics such as AUC and applying the model to large datasets.