There are three examples one might look at:

A single model method combining statistical potentials and agreement terms in a linear manner
Adding distance constraint score to QMEAN to improve local quality predictions. Evaluated are consistencies of pairwise CA-CA distances from a model with constraints extracted from homologous structures
QMEANBrane is a combination of statistical potentials targeted at local quality estimation of membrane protein models in their naturally occurring oligomeric state: after identifying the transmembrane region using an implicit solvation model, specifically trained statistical potentials get applied on the different regions of a protein model
Reference sequence (SEQRES) of submitted protein model. This sequence is used for secondary structure and solvent accessibility predictions. If not provided, the sequence gets directly extracted from the model. See the help page for further input information.
The plot relates the obtained global QMEAN4 value to scores calculated from a set of high-resolution X-ray structures.
Local quality is either estimated using the raw QMEAN scoring function or one of the two specialized functions QMEANBrane and QMEANDisCo. They all provide scores in range [0,1] with one being good.
The value of the QMEAN4 is not affected by DisCo.