Robust Inductive Matrix Completion Strategy to Explore Associations Between LincRNAs and Human Disease Phenotypes

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Long intergenic non-coding RNAs (lincRNAs) are increasingly implicated in human diseases, but many associations remain undiscovered.
• Experimental validation of lincRNA-disease links is costly and time-consuming.
• Large-scale data from HTS and GWAS offer opportunities for computational inference.

Purpose of the Study:

• To develop an in silico method for inferring lincRNA-disease associations that utilizes side information.
• To address limitations of existing Inductive Matrix Completion (IMC) methods, specifically noise, outliers, and data sparsity.
• To propose Robust Inductive Matrix Completion (RIMC) as an improved approach.

Main Methods:

• Developed Robust Inductive Matrix Completion (RIMC) using l2,1 norm loss and regularization.
• Applied RIMC to human lincRNA-OMIM disease phenotype association data.
• Integrated diverse side information for both lincRNAs and diseases.

Main Results:

• RIMC demonstrated superior performance over state-of-the-art methods in precision@k and recall@k for lincRNA disease prioritization.
• The method effectively prioritizes diseases for known lincRNAs.
• RIMC proved effective for identifying associations with novel lincRNAs and ranking new diseases for known lincRNAs.

Conclusions:

• RIMC offers a robust and effective computational approach for inferring lincRNA-disease associations.
• The method enhances disease prioritization and aids in discovering novel lincRNA-disease relationships.
• RIMC provides a valuable tool for bioinformatics research in understanding lincRNA's role in human diseases.