LPI-deepGBDT: a multiple-layer deep framework based on gradient boosting decision trees for lncRNA-protein...

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Long noncoding RNAs (lncRNAs) are crucial in biological and pathological processes.
• Identifying lncRNA-protein interactions (LPIs) is key to understanding lncRNA functions.
• Existing computational methods for LPI prediction have limitations, including prediction bias and failure to utilize diverse biological information.

Purpose of the Study:

• To develop a novel computational method for predicting unobserved lncRNA-protein interactions (LPIs).
• To overcome limitations of existing methods by integrating diverse biological data and employing a deep learning architecture.
• To improve the accuracy and efficiency of LPI prediction for both known and novel lncRNAs and proteins.

Main Methods:

• Utilized three human and two plant LPI datasets.
• Extracted biological features of lncRNAs and proteins using Pyfeat and BioProt.
• Developed a feed-forward deep architecture based on gradient boosting decision trees (LPI-deepGBDT) with dimensionally reduced and concatenated feature vectors.
• Employed forward and inverse mappings within the deep architecture for LPI prediction.

Main Results:

• LPI-deepGBDT achieved superior performance compared to five classical LPI prediction models across three cross-validation strategies.
• The model demonstrated high average AUC (0.8321, 0.6815, 0.9073) and AUPR (0.8095, 0.6771, 0.8849) values.
• Case studies identified potential interactions, including GAS5 with Q15717, RAB30-AS1 with O00425, and LINC-01572 with P35637.

Conclusions:

• The LPI-deepGBDT method effectively predicts LPIs by integrating ensemble learning and hierarchical distributed representations.
• The developed multiple-layered deep architecture enhances LPI prediction performance.
• The approach successfully probes interaction data for new lncRNAs and proteins, advancing functional genomics research.