A random forest based computational model for predicting novel lncRNA-disease associations.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Abnormal long non-coding RNA (lncRNA) regulation is linked to human diseases.
• Identifying disease-associated lncRNAs aids in understanding disease mechanisms and developing therapies.
• Existing lncRNA-disease association (LDA) prediction models often struggle with noisy and redundant data.

Purpose of the Study:

• To develop an improved model for predicting lncRNA-disease associations (LDAs).
• To address the limitations of existing models by effectively handling noisy and redundant data.
• To enhance the accuracy and reliability of LDA prediction for potential therapeutic insights.

Main Methods:

• Implemented a Random Forest and feature selection based LDA prediction model (RFLDA).
• Integrated multiple data sources: miRNA-disease associations (MDAs), LDAs, disease semantic similarity (DSS), lncRNA functional similarity (LFS), and lncRNA-miRNA interactions (LMI).
• Utilized random forest variable importance for feature selection, considering individual and joint feature effects, followed by a random forest regression model for scoring potential associations.

Main Results:

• RFLDA achieved an Area Under the Curve (AUC) of 0.976 and an Area Under the Precision-Recall Curve (AUPR) of 0.779 in 5-fold cross-validation.
• The model demonstrated superior performance compared to several state-of-the-art LDA prediction methods.
• Case studies on three cancers showed that 43 out of 45 predicted lncRNAs were experimentally validated, with the remaining two supported by other models.

Conclusions:

• The RFLDA model exhibits excellent predictive power for identifying potential disease-associated lncRNAs.
• Cross-validation and case studies confirm the model's effectiveness and reliability.
• RFLDA offers a promising tool for advancing research in lncRNA-related diseases and therapeutic development.