< Back to Article

Multiple protein-DNA interfaces unravelled by evolutionary information, physico-chemical and geometrical properties

Fig 4

Comparison of , DISPLAR, multiVORFFIP and DRNApred performance.

For , consensus predictions were obtained from 2 (in light green) and 8 (in dark green) runs out of 10. Statistical performance values are given in percentages. Top panel. HR-PDNA187 and APO82. To fairly assess DISPLAR, multiVORFFIP and DRNAPred performances, the proteins used for training these methods were removed from HR-PDNA187. We used a sequence identity cutoff of 95% and ended up with 106 (*), 87 (**) and 42 (***) proteins, respectively. Bottom panel. TEST-NABP82 and its subsets. TEST-DBP49 and TEST-DBP24 contain only DNA-binding sites, while TEST-RBP33 contains only RNA-binding sites. TEST-DBP24 and TEST-RBP33 contain only proteins sharing less than 30% sequence identity with proteins from HR-PDNA187 (see Materials and methods). DRNApred and multiVORFFIP were evaluated on their RNA-specific algorithms for the RNA-binding proteins, and on their DNA-specific ones for the DNA-binding proteins. and DISPLAR, which do not have a specific algorithm to predict RNA-binding sites, were evaluated on the same algorithm for both DNA- and RNA-binding proteins.

doi: https://doi.org/10.1371/journal.pcbi.1007624.g004