Projection-Based Neighborhood Non-Negative Matrix Factorization for lncRNA-Protein Interaction Prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Identifying long non-coding RNA (lncRNA)-protein interactions is crucial for understanding lncRNA functions.
• Experimental methods are time-consuming; computational models are increasingly used for prediction.
• Existing models often fail to integrate network topology with sequence features or predict novel interactions.

Purpose of the Study:

• To develop a computational model for predicting lncRNA-protein interactions.
• To effectively integrate biological network topology and sequence structure features.
• To improve the prediction of novel interactions, including those involving new proteins or lncRNAs.

Main Methods:

• Proposed a projection-based neighborhood non-negative matrix decomposition (PMKDN) model.
• Extracted lncRNA and protein features from sequences and expression data.
• Calculated and fused multiple lncRNA and protein similarities using GO ontology, sequences, and network information.
• Integrated similarity, feature information, and a modified interaction network into the PMKDN algorithm.

Main Results:

• PMKDN demonstrated superior performance on two benchmark datasets compared to state-of-the-art methods.
• The model effectively predicted new lncRNA-protein interactions, new lncRNAs, and new proteins.
• Case studies confirmed PMKDN's utility as an effective tool for lncRNA-protein interaction prediction.

Conclusions:

• PMKDN offers an effective approach for predicting lncRNA-protein interactions by integrating diverse biological data.
• The model shows significant potential for discovering novel molecular interactions and understanding lncRNA roles.
• PMKDN advances the field of computational prediction for RNA-protein interactions.