GAERF: predicting lncRNA-disease associations by graph auto-encoder and random forest

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying disease-related long non-coding RNAs (lncRNAs) is crucial for developing biomarkers for complex human diseases.
• Existing methods may face challenges with high-dimensional and heterogeneous biological data.

Purpose of the Study:

• To propose a novel machine learning approach, GAERF, for accurately identifying disease-related lncRNAs.
• To predict novel lncRNA-disease associations (LDAs) for improved disease understanding and management.

Main Methods:

• Constructed a heterogeneous network integrating lncRNA, miRNA, and disease relationships.
• Employed a graph auto-encoder (GAE) to learn low-dimensional representations of biological entities.
• Utilized a random forest (RF) classifier, trained on GAE features, to predict LDAs.

Main Results:

• GAE effectively reduced data dimensionality and heterogeneity, enabling accurate characterization of lncRNA-disease relationships.
• The GAERF model demonstrated superior performance in predicting LDAs compared to other methods.
• Ensemble learning within GAERF contributed to its enhanced predictive accuracy.

Conclusions:

• GAERF is an effective computational method for predicting lncRNA-disease associations.
• The approach accurately represents lncRNA-disease interactions, aiding in biomarker discovery.
• Further case studies validated GAERF's utility in identifying potential disease-related lncRNAs.