TargetSpy: a supervised machine learning approach for microRNA target prediction.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Current microRNA target site prediction algorithms predominantly rely on conserved seed matches.
• This seed match requirement may lead to the omission of a significant number of functional microRNA target sites.
• There is a need for novel computational approaches that can identify target sites irrespective of seed match conservation.

Purpose of the Study:

• To develop and evaluate TargetSpy, a novel machine learning-based computational approach for microRNA target site prediction.
• To assess TargetSpy's performance in identifying target sites with and without seed matches, including unconserved sequences.
• To compare TargetSpy's accuracy against existing algorithms across different species.

Main Methods:

• Developed TargetSpy, a machine learning model utilizing automatic feature selection.
• Incorporated a wide spectrum of compositional, structural, and base pairing features.
• Compared TargetSpy's predictions, with and without seed match filtering, against other algorithms on human and Drosophila melanogaster datasets.

Main Results:

• TargetSpy demonstrates superior performance in predicting microRNA target sites across all tested classes, including those lacking seed matches.
• The model accurately identifies between 26 and 112 functional target sites per microRNA that are missed by other algorithms.
• TargetSpy's predictions for conserved seed match classes are comparable to state-of-the-art methods.

Conclusions:

• TargetSpy significantly improves microRNA target site prediction accuracy, particularly for sites lacking seed matches.
• The tool shows broad applicability across species, performing well in human and Drosophila melanogaster after training on mouse data.
• Machine learning combined with deep sequencing data offers a powerful approach for microRNA target site prediction.