FunlncModel: integrating multi-omic features from upstream and downstream regulatory networks into a machine learning...

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Long noncoding RNAs (lncRNAs) are crucial in biological processes, but their functions are not fully understood.
• Existing methods often overlook upstream epigenetic regulators when predicting lncRNA function.
• Leveraging multi-omic regulatory networks can enhance the prediction of lncRNA roles.

Purpose of the Study:

• To develop FunlncModel, an interpretable machine learning framework for predicting functional lncRNAs.
• To integrate diverse (epi)genetic and functional genomic features from regulatory networks.
• To identify novel functional lncRNAs and improve understanding of their roles in cellular processes.

Main Methods:

• Utilized a random forest algorithm to analyze over 2000 datasets across 11 data types.
• Integrated features including transcription factors, histone modifications, enhancers, methylation sites, and mRNAs.
• Developed a computational framework integrating upstream/downstream multi-omic regulatory networks.

Main Results:

• FunlncModel achieved high classification performance in human embryonic stem cells (AUROC=0.95, AUPRC=0.97).
• The model successfully identified known and discovered novel functional lncRNAs.
• Validated efficacy across 27 cancer-related functional prediction tasks, highlighting epigenetic features as strong predictors.

Conclusions:

• FunlncModel is a robust and stable prediction model for identifying functional lncRNAs in specific cellular contexts.
• Epigenetic regulatory features, such as transcription factors and histone modifications, are key predictors of lncRNA function.
• The framework provides a valuable resource for lncRNA research in molecular biology and disease.