Computationally predicting protein-RNA interactions using only positive and unlabeled examples.

Area of Science:

• Computational biology
• Bioinformatics
• Molecular biology

Background:

• Protein-RNA interactions (PRIs) are crucial for numerous cellular processes, including gene regulation and antiviral defense.
• High-throughput technologies generate vast amounts of PRI data, necessitating efficient computational prediction methods.
• Existing computational PRI prediction methods often rely on artificially generated negative samples, limiting their accuracy.

Purpose of the Study:

• To develop a novel computational method for predicting Protein-RNA interactions (PRIs) using only positive and unlabeled data.
• To address the limitations of existing methods that use artificial negative samples.
• To improve the accuracy and efficiency of computational PRI prediction.

Main Methods:

• Developed PRIPU, a method employing biased-support vector machine (SVM) for PRI prediction using positive and unlabeled examples.
• Extracted sequence-based features to represent PRIs and employed feature selection to reduce dimensionality.
• Introduced Explicit Positive Recall (EPR) as a novel performance measure for evaluating learning from positive and unlabeled data.

Main Results:

• PRIPU demonstrated superior performance compared to four existing computational methods across three benchmark datasets.
• The method successfully predicted previously unknown Protein-RNA interactions.
• The proposed EPR metric is effective for evaluating models trained on positive and unlabeled data.

Conclusions:

• PRIPU offers a significant advancement in computational PRI prediction by effectively utilizing positive and unlabeled data.
• The method provides a more accurate and reliable approach to identifying PRIs compared to existing techniques.
• PRIPU has the potential to accelerate research in molecular biology and disease mechanisms involving PRIs.