tREMAP

tREMAP predicts target genes of transcription factors (TFs) by integrating known gene–TF associations and protein–protein interaction networks using a one-class collaborative filtering approach.

Key Features:

• Algorithmic foundation: Implements a one-class collaborative filtering algorithm based on regularized, weighted nonnegative matrix tri-factorization to predict unobserved TF–gene associations.
• Data integration: Integrates known gene–TF association data with protein–protein interaction (PPI) networks to inform predictions.
• Benchmark performance: Demonstrates superior performance relative to REMAP (bi-factorization) across AUC, MAP, MPR, and HLU metrics.
• Quantitative evaluation: Achieved 37.8% accuracy for the top 495 predicted associations with a 4.19-fold enrichment over random expectation on an independent dataset.
• Validation of novel predictions: Many novel TF–gene associations predicted by tREMAP have supporting evidence in the literature.
• Versatility: Applicable to canonical TF–target datasets and adaptable to tissue-specific datasets and integration of multiple omics datasets.

Scientific Applications:

• Gene regulatory network inference: Prioritizes candidate TF–gene regulatory interactions for incorporation into gene regulatory network models.
• Integrative omics analysis: Enhances prediction of TF targets by combining TF–target associations with PPI and other omics datasets.
• Experimental prioritization: Ranks TF–gene associations to guide targeted experimental validation and hypothesis generation in studies of cellular processes and disease pathogenesis.

Methodology:

Integrates known gene–TF associations and protein–protein interaction networks and applies a regularized, weighted nonnegative matrix tri-factorization one-class collaborative filtering algorithm, with performance assessed using AUC, MAP, MPR, HLU and independent-dataset accuracy/enrichment metrics.