Structure-based learning to predict and model protein-DNA interactions and transcription-factor co-operativity in...

Area of Science:

• Genomics
• Structural Biology
• Computational Biology

Background:

• Transcription factor (TF) binding is crucial for genomic regulation.
• Experimental methods for TF-DNA binding specificity are costly and laborious, leaving many TF binding preferences unknown.
• Current computational methods struggle with remote homology in predicting TF binding.

Purpose of the Study:

• To develop a structure-based learning approach for predicting TF binding preferences.
• To enable automated modeling of TF regulatory complexes.
• To overcome limitations of existing methods, especially for TFs with unknown binding data.

Main Methods:

• A structure-based learning approach is used to predict TF binding preferences as motifs.
• Predicted motifs are scanned against DNA sequences to identify potential binding sites.
• Binding scores are assigned, and higher-order regulatory complexes are modeled structurally, including protein-protein interactions.

Main Results:

• The approach successfully predicts TF binding preferences and models TF-DNA and TF-TF interactions.
• Demonstrated advantage over nearest-neighbor prediction in remote homology scenarios.
• Successfully modeled known complexes like the interferon-β enhanceosome and pluripotency factor complexes.

Conclusions:

• The developed structure-based learning method offers an efficient and accurate way to predict TF binding and model regulatory complexes.
• This approach can help fill gaps in knowledge for uncharacterized TFs.
• Enables deeper understanding of gene regulation through automated complex modeling.