Performance of various scoring functions.

Scoring functions are FTdock's Surface Complementarity Score (SCS), electrostatics via Coulomb's law, a statistical potential based on contacts (C-S), a distance-dependent statistical potential (D-S) and our work. Each scoring function is used in scoring three sets of proteins; the set of N = 47 known DNA-binding proteins (set K), the set of proteins with pI<7, and the set with pI>7. From the distribution of scores on a given set, say set K, we obtain the average score μ and the standard deviation σ of scores: this in turn allows locating any scoring value χ following a Z-score computation Z = (χ−μ)/σ. We report here the Z-scores of the median scores on set pI<7 and set pI>7. Positive Z-Score values indicate that the median has a value greater than the mean of set K, while negative ones lower. Ideally, the mean of set K should be lower than the median of the other sets (i.e. set K has better scores). The greater the Z-Score the better is the scoring function at discriminating DNA-binding proteins from non-binders. Normalized overlapping areas are also reported.