Table 4.

Description of the methods for evaluating protein model quality that were compared in the benchmark

Method	Description (used measure)
1. ProQ-LG	neural network trained to predict LGscore
2. ProQ-MX	neural network trained to predict MaxSub
3. Errat (Colovos and Yeates 1993)	statistics of interactions between different atom types (Total-Errat)
4. Prosa II (Sipple 1993b)	statistical potential, (zp surf, polyprotein pII3.0.long.ply)
5. Verify3D (Eisenberg et al. 1997)	3D–1D profiles, gives a score for each residue (average score)