WINTF

WINTF predicts unobserved transcription factor–gene associations by applying weighted imputation, neighborhood-regularized tri-factorization, and one-class collaborative filtering to noisy, sparse regulatory data.

Key Features:

• Weighted Imputation: Incorporates a weighting mechanism that accounts for varying confidence in known TF-gene associations to mitigate bias in the input data.
• Neighborhood Regularization: Integrates protein-protein interaction networks for neighborhood regularization to leverage biological context in predictions.
• Tri-factorization Methodology: Employs tri-factorization to decompose the association matrix into three matrices that capture complex TF–gene relationships.
• One-class Collaborative Filtering: Applies a one-class collaborative filtering framework tailored for predicting associations without requiring negative examples.

Scientific Applications:

• TF–gene interaction prediction: Prediction of unobserved transcription factor targets to support identification of regulatory relationships.
• Gene regulatory network analysis: Reconstruction and analysis of gene regulatory networks from sparse and noisy TF-gene association data.
• Benchmarking and validation: In benchmark studies WINTF achieved 37.8% accuracy on the top 495 predicted associations with an enrichment factor of 4.19 over random, and many predictions were corroborated by independent literature.

Methodology:

Uses known but noisy and incomplete TF-gene association data and protein-protein interaction networks, applying weighted imputation, neighborhood regularization, tri-factorization, and one-class collaborative filtering to predict unobserved associations.