LDNFSGB

LDNFSGB predicts associations between long non-coding RNAs (lncRNAs) and human diseases by integrating network feature similarity measures and gradient boosting to identify potential lncRNA–disease links.

Key Features:

• Comprehensive feature vector: Integrates Disease Semantic Similarity (DISSS), lncRNA Function Similarity (LNCFS), lncRNA Gaussian Interaction Profile Kernel Similarity (LNCGS), Disease Gaussian Interaction Profile Kernel Similarity (DISGS), and lncRNA–disease interaction (LNCDIS) into a single feature representation.
• Dual methodology for DISSS and LNCFS: Employs two distinct methods to compute DISSS and LNCFS, capturing both local and global information for disease semantics and lncRNA functions.
• Dimensionality reduction via autoencoder: Uses an autoencoder to reduce feature vector dimensionality and extract optimal feature parameters from the original dataset.
• Gradient boosting prediction: Applies a gradient boosting algorithm to predict lncRNA–disease associations.

Scientific Applications:

• Candidate prioritization: Prioritizes lncRNAs for experimental validation in studies of human diseases, including cancers and non-cancer conditions.
• In silico screening: Provides rapid computational screening to complement time-consuming experimental detection methods for lncRNA–disease association discovery.
• Case study validation: Demonstrated applicability through case studies focusing on six diseases.

Methodology:

Constructs a feature vector comprising DISSS, LNCFS, LNCGS, DISGS, and LNCDIS with two methods for DISSS/LNCFS, applies an autoencoder for dimensionality reduction, uses gradient boosting for prediction, and evaluates performance on three public datasets via hold-out, leave-one-out, and ten-fold cross-validation.