lDDT measures how well the environment in a reference structure is reproduced in a protein model. It is computed over all pairs of atoms in the reference structure at a distance closer than a predefined threshold Ro (called inclusion radius), and not belonging to the same residue. These atom pairs define a set of local distances L. A distance is considered preserved in the model M if it is, within a certain tolerance threshold, the same as the corresponding distance in . If one or both the atoms defining a distance in the set are not present in M, the distance is considered non-preserved. For a given threshold, the fraction of preserved distances is calculated. The final lDDT score is the average of four fractions computed using the thresholds 0.5 Å, 1 Å, 2 Å and 4 Å, the same ones used to compute the GDT-HA score (Battey et al., 2007 (link)). For partially symmetric residues, where the naming of chemically equivalent atoms can be ambiguous (glutamic acid, aspartic acid, valine, tyrosine, leucine, phenylalaine and arginine), two lDDTs, one for each of the two possible naming schemes, are computed using all non-ambiguous atoms in M in the reference. The naming convention giving the higher score in each case is used for the calculation of the final structure-wide lDDT score.
The lDDT score can be computed using all atoms in the prediction (the default choice), but also using only distances between Cα atoms, or between backbone atoms. Interactions between adjacent residues can be excluded by specifying a minimum sequence separation parameter. Unless explicitly specified, the calculation of the lDDT scores for all experiments described in this article has been performed using default parameters, i.e. Ro = 15 Å, using all atoms at zero sequence separation.