Target Gene Prediction of Transcription Factor Using a New Neighborhood-regularized Tri-factorization One-class...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Understanding genome-wide transcription factor (TF) target gene profiles is crucial for deciphering transcriptional regulation.
• Experimental methods for identifying TF targets are often costly and complex, limiting comprehensive analysis.
• Computational prediction offers a viable approach to infer unobserved TF-gene associations.

Purpose of the Study:

• To develop and evaluate tREMAP, a novel computational algorithm for predicting transcription factor target genes.
• To assess the performance of tREMAP against existing methods using benchmark and independent datasets.
• To provide a tool for advancing the study of gene regulatory networks.

Main Methods:

• Developed tREMAP, a one-class collaborative filtering algorithm utilizing regularized, weighted nonnegative matrix tri-factorization.
• Integrated known gene-TF associations and protein-protein interaction networks.
• Benchmarked tREMAP against REMAP (a bi-factorization algorithm) and evaluated on independent datasets.

Main Results:

• tREMAP significantly outperformed REMAP in transcription factor target gene prediction across four key performance metrics (AUC, MAP, MPR, HLU).
• Achieved 37.8% prediction accuracy on the top 495 predicted associations from independent datasets, with an enrichment factor of 4.19.
• Validated numerous predicted novel TF-gene associations through literature evidence.

Conclusions:

• tREMAP is a highly effective computational tool for predicting transcription factor target genes, surpassing existing methods.
• The algorithm demonstrates potential for integration with various omics data and application to tissue-specific datasets.
• tREMAP offers a valuable framework for advancing the understanding of gene regulatory networks.