Native recognition performance comparison among LoCo, our ω-only function and DFMAC both with and without its ω component is shown. All reported measures are averages over the 77 decoy sets in the final testing group. Lower scores are better for Ranknat and RMSDbest. Higher ones are better for Znat, CCnat and FEnat. The ω-only function is able to pick out native structures quite well, but when it fails, its choices are essentially random. In the two measures for which the ω-only function does poorly (CCnat and FEnat), DFMAC performance improves when its ω component is removed. All metrics are defined in Performance measures at the end of Methods.