PreCisIon: PREdiction of CIS-regulatory elements improved by gene's positION

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Traditional methods for predicting transcription factor (TF) target genes rely on TF binding site motifs, often represented as position-specific scoring matrices.
• These motifs have limitations in accuracy and availability, leading to unreliable predictions of target genes.

Purpose of the Study:

• To improve the accuracy of TF binding site prediction by integrating genome layout information with sequence data.
• To develop a machine learning algorithm that adaptively combines local sequence and global gene position information for robust target gene identification.

Main Methods:

• A machine learning algorithm, PreCisIon, was developed to combine traditional sequence information with novel genome layout data (gene proximity and periodic spacing).
• PreCisIon adaptively combines weak classifiers based on local binding sequences and global gene positions to build a strong gene target classifier.
• Cross-validation analysis was performed on 20 major TFs from Bacillus subtilis and Escherichia coli.

Main Results:

• PreCisIon consistently improves upon methods relying solely on sequence information.
• For Bacillus subtilis and Escherichia coli, PreCisIon achieved average areas under the receiver operating characteristic curve of 70% and 60%, respectively.
• The method demonstrated average sensitivities of 80% and 70%, and specificities of 60% and 56% for the respective organisms.

Conclusions:

• The integration of genome layout information significantly enhances the prediction of TF target genes.
• PreCisIon offers a flexible framework for incorporating future advances in gene target prediction criteria.
• Newly predicted gene targets were validated through functional consistency analysis, including Gene Ontology enrichment and literature review.