MicroRNA target prediction using thermodynamic and sequence curves.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) regulate gene expression by binding to mRNA targets.
• Existing computational methods often use scalar features and focus on canonical miRNA-mRNA interactions.
• Novel approaches are needed to capture complex interaction profiles and non-canonical binding sites.

Purpose of the Study:

• To develop a machine learning model, Avishkar, for improved miRNA target prediction.
• To incorporate spatial profiles of miRNA-mRNA interactions using B-spline curves.
• To uniformly model both canonical and non-canonical seed matches.

Main Methods:

• Utilized smooth B-spline curves to represent spatial profiles of input features (thermodynamic, sequence).
• Developed a novel seed enrichment metric to model canonical and non-canonical seed matches.
• Trained a linear Support Vector Machine (SVM) model using experimental CLIP-seq data.

Main Results:

• Demonstrated high enrichment values for conserved seed-match patterns across species.
• Showed that a majority of miRNA binding sites involve non-canonical matches.
• Achieved superior performance over established methods for both canonical and non-canonical sites, outperforming them significantly.

Conclusions:

• Developed an efficient SVM-based model (Avishkar) for miRNA target prediction with superior performance (approx. 20% improvement).
• The model demonstrates high accuracy across different species and target types (canonical/non-canonical).
• Introduced the first distributed framework for miRNA target prediction using Apache Hadoop and Spark.