Long Noncoding RNA function prediction via multiview cross-contrastive learning combined with multiscale semantic...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Long noncoding RNAs (lncRNAs) are crucial regulators in complex diseases, but predicting their functions is challenging.
• Existing methods struggle with feature extraction from multimodal omics data and modeling gene ontology (GO) structures.

Purpose of the Study:

• To develop a novel framework, MiCLSAO, for enhanced lncRNA function prediction.
• To improve the extraction of discriminative features from omics data and the modeling of GO's semantic and topological information.

Main Methods:

• MiCLSAO employs multiview cross-contrastive learning with attention for lncRNA feature extraction from omics similarity networks.
• Graph convolutional networks refine GO term representations using multiscale topological and semantic relationships.
• A Kolmogorov-Arnold network integrates lncRNA and GO term representations for prediction.

Main Results:

• MiCLSAO significantly outperforms state-of-the-art methods across multiple evaluation metrics.
• The framework demonstrates strong capabilities in recovering known lncRNA functions and identifying novel ones.
• Experimental results highlight MiCLSAO's practical utility in providing informative lncRNA annotations.

Conclusions:

• MiCLSAO offers a robust and effective approach for lncRNA function prediction.
• The framework advances the understanding of lncRNA roles in disease biology.
• MiCLSAO has the potential to accelerate therapeutic development for complex diseases.