BiWalkLDA

BiWalkLDA predicts associations between long non-coding RNAs (lncRNAs) and diseases by applying a bi-random walk on networks integrating interaction profiles and gene ontology information.

Key Features:

• Bi-Random Walk Methodology: BiWalkLDA employs a bi-random walk to traverse the constructed lncRNA-disease network and addresses the cold-start problem by leveraging neighbors' interaction profile information.
• Integration of Interaction Profiles and Gene Ontology: The algorithm constructs networks by combining interaction profiles with gene ontology data to inform association predictions.
• Parameter Optimization: BiWalkLDA includes mechanisms to analyze the effects of parameters α, β, l, and r to tailor and optimize algorithm performance.
• Performance Evaluation: The method was tested on three biological datasets containing 528 lncRNAs, 545 diseases, and 1216 interactions and showed superior accuracy and specificity compared to SIMCLDA, LDAP, and LRLSLDA using leave-one-out validation.
• Case Study Validation: In a prostate cancer case study, eight of the top ten predicted disease-related lncRNAs were previously confirmed in the literature.

Scientific Applications:

• lncRNA-disease association prediction: Identification of candidate disease-related lncRNAs for downstream experimental validation.
• Molecular mechanism exploration: Prioritization of lncRNAs to support investigations into molecular mechanisms underlying diseases.
• Non-coding RNA bioinformatics: Integration of interaction profile and gene ontology data to support computational studies in non-coding RNA biology.

Methodology:

The framework constructs a network by integrating interaction profiles with gene ontology information, applies a bi-random walk that leverages neighbors' interaction profile information to address the cold-start problem, analyzes parameters α, β, l, and r, and uses leave-one-out validation on three datasets (528 lncRNAs, 545 diseases, 1216 interactions) with a prostate cancer case study for evaluation.