SRG-vote

SRG-vote predicts microRNA (miRNA)-gene interactions using multiple embedding methods and deep learning to characterize regulatory relationships relevant to precision medicine and tumor heterogeneity.

Key Features:

• Embedding Techniques: Uses doc2vec for sequential embeddings from molecular sequences and role2vec, Graph Convolutional Networks (GCN), and Gaussian Mixture Models (GMM) for geometrical embeddings from similarity- and pairwise-based network data.
• Deep Learning Models: Employs Long Short-Term Memory (LSTM) and Bidirectional LSTM (Bi-LSTM) networks to model patterns in the extracted embeddings.
• Ensemble Voting: Aggregates outputs from multiple models and data sources via a voting system to improve prediction performance measured by metrics such as AUC.
• Input Data Types: Processes raw molecular sequences and network-based datasets derived from similarity and pairwise relationships.

Scientific Applications:

• miRNA–gene interaction prediction: Predicts regulatory relationships between microRNAs and genes to elucidate gene regulation mechanisms.
• Cancer research and precision medicine: Supports identification of candidate therapeutic targets and biomarkers for cancer, addressing tumor heterogeneity.
• Molecular genetics studies: Provides computational predictions to assist investigations of gene regulation and molecular mechanisms.

Methodology:

Computational steps include data preparation of high-throughput molecular sequences and network datasets, feature extraction using doc2vec, role2vec, GCN, and GMM, modeling with LSTM and Bi-LSTM networks, and integration of model outputs via an ensemble voting system.