Multitask learning model for predicting non-coding RNA-disease associations: Incorporating local and global context.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Long non-coding RNAs (lncRNAs) and microRNAs (miRNAs) are key non-coding RNAs implicated in various diseases.
• Current computational methods often analyze lncRNA-miRNA-disease associations in isolation, leading to limited predictive power and generalizability.
• Higher-order topological information in ncRNA-disease relationships is frequently overlooked by existing approaches.

Purpose of the Study:

• To develop an integrated computational model for simultaneously predicting lncRNA-disease associations, miRNA-disease associations, and lncRNA-miRNA interactions.
• To address the limitations of isolated task analysis and incorporate higher-order topological information in ncRNA-disease relationship prediction.
• To enhance the accuracy and generalizability of predicting non-coding RNA (ncRNA) involvement in diseases.

Main Methods:

• Proposed the Multi-Task Learning for Non-coding RNA-Disease Association (MTL-NRDA) model, a multi-task learning framework.
• Integrated multi-source information using a heterogeneous network of lncRNAs, miRNAs, and disease association/similarity networks.
• Employed higher-order graph convolutional networks (HOGCN) for local feature aggregation and a transformer encoder for global feature extraction to optimize node embeddings.

Main Results:

• MTL-NRDA demonstrated superior performance compared to existing models on two independent datasets.
• Ablation studies validated the effectiveness of individual model components and the multi-task learning strategy.
• Case studies on breast and liver cancers highlighted the model's practical applicability in identifying disease-related ncRNA associations.

Conclusions:

• The MTL-NRDA model effectively predicts lncRNA-disease and miRNA-disease associations by leveraging a multi-task learning framework and integrating diverse data sources.
• The model's ability to capture both local and global topological features enhances the prediction of ncRNA-disease relationships.
• MTL-NRDA offers a promising computational tool for advancing diagnostic, preventive, and therapeutic strategies for ncRNA-associated diseases.