Each model is compared pair-wise with all other models using the LDDT and LGA packages.

On the global scale, the similarity between each pair of models is quantified by the overall LDDT or GDTTS score. The cumulative similarity of a model to all other models (consensus score) is calculated as an average of all pair-wise scores for this model. The consensus LDDT_cons and GDTTS_cons scores are reported in columns 5 and 7 of the results table. Higher consensus scores do not necessarily signify better models, but rather those that are more similar to others.

Models are clustered on each of the targets using single-linkage hierarchical clustering with the inter-model distance equal to (1 - model_pairwise_score). The clustering dendrograms according to LDDT and CAD-scores help identify groups of similar models (credit to Kliment Olechnovic, Vilnius University).

On the local scale, the similarity between local regions in each pair of models is quantified by the per-residue LDDT or S-score. The local LDDT score reports similarity of distance patterns for a selected residue in two compared models. The S-score is a measure of proximity of corresponding residues in the LGA sequence-dependent superposition; in essence, it is the distance d between CA atoms of the corresponding residues transformed to [0,1] range using the S-function=1/(1+(d/5)^2). The local consensus score for a residue in a model is the average of local LDDT scores or S-scores for this residue from all pair-wise LDDT or LGA runs. The per-residue consensus scores are illustrated in the colored bar plots - left for the LDDT, right for the S-score. The residue numbering is shown in the title of the graphic columns 4 and 6, the local score color-coding is provided above the results table.

Clicking on a bar opens a window with rendering of the selected model colored according to the underlying bar scheme.