A method for discriminating native protein-DNA complexes from decoys using spatial specific scoring matrices

Decoding protein-DNA interactions is important to understanding gene regulation and has been investigated by worldwide scientists for a long time. However, many aspects of the interactions still need to be uncovered. The crystal structures of protein-DNA complexes reveal detailed atomic interactions between the proteins and DNA and are an excellent resource for investigating the interactions. In this study, we profiled the spatial distribution of protein atoms around six structural components of the DNA, which are the four bases, the deoxyribose sugar and the phosphate group. The resultant profiles not only revealed the preferred atomic interactions across the protein-DNA interface but also captured the spatial orientation of the interactions. The profiles are a useful tool for investigating protein-DNA interactions. We tested the strength of profiles in two experiments, discrimination of native protein-DNA complexes from decoys with mutant DNA and discrimination of native protein-DNA complexes from near-native docking decoys. The profiles achieved an average Z-score of 7.41 and 3.22 respective on benchmark datasets for the tests, both are better than other knowledge-based energy functions that model protein-DNA interaction based on atom pairs.