Constructing discriminative feature space for LncRNA-protein interaction based on deep autoencoder and marginal...

Area of Science:

• Computational Biology
• Molecular Biology
• Bioinformatics

Background:

• Long non-coding RNAs (lncRNAs) regulate protein functions in essential biological processes.
• Accurate identification of lncRNA-protein interactions is vital for elucidating molecular mechanisms.
• Existing machine learning methods face challenges in constructing discriminative feature spaces for these interactions.

Purpose of the Study:

• To develop a novel computational method for accurately predicting lncRNA-protein interactions.
• To address the limitations of current methods in feature space construction for lncRNA-protein interactions.

Main Methods:

• A novel method, DFRPI, was developed using deep autoencoder and marginal Fisher analysis.
• Initial features were extracted from lncRNA and protein sequences and structures.
• A deep autoencoder learned precise interaction descriptions, optimized by marginal Fisher analysis for a discriminative feature space.
• A random forest predictor was trained on the optimized feature space.

Main Results:

• The DFRPI predictor achieved high performance metrics: precision (0.920), recall (0.916), accuracy (0.918), MCC (0.836), specificity (0.920), sensitivity (0.916), and AUC (0.906).
• The proposed method demonstrated superior performance compared to existing approaches for lncRNA-protein interaction prediction.
• DFRPI effectively generates a discriminative feature space for accurate interaction detection.

Conclusions:

• The DFRPI method provides a robust and effective approach for predicting lncRNA-protein interactions.
• The developed feature space is crucial for distinguishing interactions accurately.
• The findings contribute to a better understanding of lncRNA-protein interactions and their biological significance.