ComBind

ComBind predicts binding specificities of transcription factor (TF) pairs on DNA to model cooperative TF–TF–DNA interactions.

Key Features:

• Random Forest Methodology: ComBind employs random forest algorithms to model complex TF–TF–DNA interaction patterns.
• JointRF Model: JointRF builds upon position weight matrices (PWMs) using sub-sequences from large-scale CAP-SELEX DNA libraries and achieves AUROC 0.75 compared to 0.59 for orientation- and spacing-specific pairwise PWMs.
• ComBind Model: The ComBind model simultaneously considers multiple orientations and spacings between two transcription factors without requiring prior knowledge of binding preferences and attains AUROC 0.78 (p<0.00195).
• CAP-SELEX Training Data: ComBind is trained on large-scale CAP-SELEX DNA libraries containing sequences enriched for binding by specific TF pairs.

Scientific Applications:

• Enhanced Prediction Accuracy: ComBind improves prediction of TF–TF binding sites relative to orientation- and spacing-specific pairwise PWMs and the JointRF approach.
• Flexible Binding Models: ComBind enables modeling of TF pair interactions across multiple orientations and spacings, facilitating study of flexible cooperative binding.

Methodology:

ComBind trains random forest models on CAP-SELEX DNA libraries; JointRF derives features from PWMs using CAP-SELEX sub-sequences; the ComBind model evaluates multiple orientations and spacings between TFs without prior binding-preference specification.