Abstract

Background: Long non-coding RNAs (lncRNAs) are nonprotein-coding transcripts of more than 200 nucleotides in length. In recent years, studies have shown that long non-coding RNAs (lncRNA) play a vital role in various biological processes, complex disease diagnosis, prognosis, and treatment.

Objective: Analysis of known lncRNA-disease associations and prediction of potential lncRNA-disease associations are necessary to provide the most probable candidates for subsequent experimental validation.

Methods: In this paper, we present a novel robust computational framework for lncRNA-disease association prediction by combining the ℓ1-norm graph with multi-label learning. Specifically, we first construct a set of similarity matrices for lncRNAs and diseases using known associations. Then, both lncRNA and disease similarity matrices are adaptively re-weighted to enhance the robustness via the ℓ1- norm graph. Lastly, the association matrix is updated with a graph-based multi-label learning framework to uncover the underlying consistency between the lncRNA space and the disease space.

Results: We compared the proposed method with the four latest methods on five widely used data sets. The experimental results show that our method can achieve comparable performance in both five-fold cross-validation and leave-one-disease-out cross-validation prediction tasks. The case study of prostate cancer further confirms the practicability of our approach in identifying lncRNAs as potential prognostic biomarkers.

Conclusion: Our method can serve as a useful tool for the prediction of novel lncRNA-disease associations.

Keywords: lncRNA-disease association, similarity matrices, ℓ1-norm graph, multi-label learning, prostate cancer, protein.

Graphical Abstract