Iridium

Iridium [Warren-2012] is a metric used to estimate the model quality of a structure resulting from an X-ray crystallography experiment. The metric evaluates what we deem essential considerations for using protein-ligand structures in drug discovery, particularly virtual screening.

The metric was developed by aggregating previously published docking validation sets, evaluating each of them by eye and even re-refining some of the data. The high quality structures were then deposited into a new dataset called Iridium HT [Warren-2012].

The metric takes a number of parameters into account and categorizes structures into 4 different categories, grading their trustworthiness to use in modeling.

  • HT - Highly Trustworthy
  • MT - Mildly Trustworthy
  • NT - Not Trustworthy
  • NA - Not Applicable

The latter category is used when no electron density data is available, or when the structure being evaluated is not coming from an X-ray crystallography experiment. Electron density data is available for at least 85% of the public PDB data and is required for all new depositions. Historical data is also being recovered, so that percentage is only expected to increase.

Note

Regarding applicability:

  • Iridium is not currently an appropriate metric for electron microscopy (EM) experiments, in part because the contour level, which we use as 1 sigma in X-ray experiments, is determined on a per experiment basis, and also in part due to different diffraction behavior of heavy atoms comparing to the X-ray.
  • Iridium is not currently an appropriate metric for neutron diffraction experiments, due to the different diffraction behavior for heavy atoms (e.g. sulfur and phosphorus are largely invisible in neutron diffraction data).
  • Iridium relies on a bound ligand, and is therefore not appropriate for apo structures.

Categorization

The initial Iridium categorization was done by eye. The rules developed were then formalized to be able to rigorously assess the Iridium category on every X-ray structure. In some cases the original publication is not explicit about threshold values, but the metric below follows the paper in spirit and is approved by the original authors. The outcome of formalizing the rules, resulted in minor category discrepancies on the published dataset, is likely due to the more stringent adherence to the rules.

The criteria to consider include both global and local quality metrics and are as follows:

  • DPI - The diffraction-component precision index or global precision estimate [Cruickshank-1999]
  • R-free value
  • Crystallographic Resolution
  • Density coverage of the ligand heavy atoms
  • Density coverage of the active site residue heavy atoms (including co-factors)
  • Occupancy of ligand heavy atoms
  • Occupancy of active site heavy atoms
  • Alternate locations of the ligand
  • Alternate locations of the active site residues
  • Presence of crystal packing residues near binding site
  • Presence of excipients near binding site
  • Whether ligand is covalently bound

First, the initial consideration is based on the density coverage of the ligand and the active site residues.

  HT MT NT
Ligand > 0.90 < 0.90 and > 0.50 < 0.50
Active Site > 0.95 < 0.95 and > 0.50 < 0.50

Subsequently, a structure can be demoted from HT to MT, if any of the conditions below are true:

  • DPI > 0.50
  • Presence of alternate locations of the ligand or the active site residues
  • Any ligand heavy atoms with an occupancy < 0.90
  • Any active site heavy atoms with an occupancy < 0.50
  • Perception of interactions between packing residues and the ligand using OEPerceiveInteractionOptions with default settings.
  • Perception of interactions between any excipient and the ligand using OEPerceiveInteractionOptions with default settings.
  • Perception of covalent interactions of the ligand in the active site using OEPerceiveInteractionOptions with default settings.

Furthermore, any structure will be demoted to NT if the Rfree value is larger than 0.45 and the resolution is below 3.5A, as we consider that an irrational R-free value.

Note

Not all structures report a DPI, and in such cases it is calculated using OECalculateDPI.

See also