A machine learning framework that integrates multi-omics data predicts cancer-related LncRNAs

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Long non-coding RNAs (lncRNAs) are key in cancer diagnosis and prognosis.
• Understanding lncRNA-disease mechanisms is challenging.
• Multi-omics data and machine learning offer new avenues for research.

Purpose of the Study:

• To develop a machine learning framework for predicting disease-related lncRNAs.
• To leverage multi-omics data and neural network methods for improved association prediction.
• To identify novel cancer-related lncRNAs.

Main Methods:

• Proposed LGDLDA (LncRNA-Gene-Disease association networks based LncRNA-Disease Association prediction) framework.
• Calculated similarity matrices for lncRNAs, genes, and diseases using various data types (expression, interactions, ontologies).
• Integrated neighborhood information via neural network nonlinear feature learning and node representations.

Main Results:

• LGDLDA demonstrated improved performance in cancer-related lncRNA prediction compared to existing methods.
• The framework showed superior stability and accuracy in predicting disease-related lncRNAs across simulation and real cancer datasets.
• Successfully applied to gastric, colorectal, and breast cancer data.

Conclusions:

• LGDLDA effectively predicts disease-related lncRNAs by integrating critical information.
• The method offers a robust and accurate approach for lncRNA-disease association prediction.
• Identified potential cancer-related lncRNAs for further investigation.