MiRTif: a support vector machine-based microRNA target interaction filter.

Area of Science:

• Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• MicroRNAs (miRNAs) are crucial gene regulators that bind to target genes, but identifying these interactions is challenging.
• Existing bioinformatics tools predict numerous miRNA targets but often yield high false positive rates.
• Machine learning offers a promising approach for filtering predicted miRNA targets using experimentally validated data.

Purpose of the Study:

• To develop and evaluate MiRTif, a machine learning-based system for filtering microRNA:target interactions.
• To improve the accuracy of miRNA target prediction by reducing false positives.
• To provide a post-processing tool for existing miRNA target prediction software.

Main Methods:

• Developed MiRTif, a support vector machine (SVM) classifier.
• Trained the SVM with 195 experimentally validated positive and 38 negative miRNA:target interaction pairs.
• Utilized k-gram frequencies from seed, non-seed, and entire interaction regions as features.
• Selected informative features based on discriminating abilities and assessed performance using 10-fold cross-validation.

Main Results:

• MiRTif achieved an Area Under the ROC Curve (AUC) of 0.86, with 83.59% sensitivity and 73.68% specificity.
• The system successfully identified the majority of false positive miRNA:target interactions (28 out of 38).
• Investigated potential over-fitting issues with the negative sample set.

Conclusions:

• MiRTif functions as a post-processing filter for miRNA target prediction software.
• It effectively distinguishes likely real miRNA:target interactions from pseudo ones.
• The MiRTif system is accessible online for researchers.