A novel lncRNA-protein interaction prediction method based on deep forest with cascade forest structure.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Long noncoding RNAs (lncRNAs) are crucial regulators of biological processes through interactions with RNA-binding proteins.
• Accurate identification of lncRNA-protein interactions (LPIs) is vital for understanding lncRNA functions and mechanisms.
• Current computational LPI prediction methods suffer from dataset bias, limited discovery of novel interactions, and suboptimal performance.

Purpose of the Study:

• To develop an advanced computational method for predicting lncRNA-protein interactions (LPIs).
• To overcome limitations of existing methods, including prediction bias and inability to identify novel interactions.
• To enhance the accuracy and scope of LPI prediction for better functional characterization of lncRNAs.

Main Methods:

• A Deep Forest-based LPI prediction method (LPIDF) was developed.
• Features for lncRNAs and proteins were constructed using four-nucleotide composition and BioSeq2vec with an encoder-decoder structure.
• A cascade forest structure within the Deep Forest model was employed for LPI prediction.

Main Results:

• LPIDF demonstrated superior performance compared to four classical association prediction models across three fivefold cross-validations.
• The method achieved high average AUCs (0.9012, 0.6937, 0.9457) and AUPRs (0.9022, 0.6860, 0.9382).
• LPIDF successfully predicted a potential interaction between lncRNA FTX and protein P35637, requiring further experimental validation.

Conclusions:

• The developed LPIDF method offers a robust and accurate approach for LPI prediction.
• LPIDF effectively addresses dataset bias and enhances the discovery of novel lncRNA-protein interactions.
• The findings highlight the potential of deep learning models in advancing the study of lncRNA biology.