Comparing your docking results to a known reference structure

The comparison to a reference structure has been streamlined, and can now be made simply by using the [caprieval] module. This module will compute CAPRI criteria, of all the generated structures with respect to a reference one, with ligand-RMSD (l-RMSD), interface-ligand-RMSD (il-RMSD), Fraction of Native contacts (Fnat), DockQ and global-RMSD (RMSD) metrics. This is also extended to the cluster level, enabling to rank clusters.

See here the full documentation related to the [caprieval] module.

Here is a schematic example of how to use the [caprieval] module:

# Some previous modules in the workflow
# ...

# Use CAPRIeval to compare previously generated models to a reference
[caprieval]
reference_fname = "target_complex.pdb"

# Some more modules until the end of the workflow
# ...

Notes:

  • The best scoring complex will be used as a reference if the reference_fname in the [caprieval] module is not specified.
  • If a clustering step is placed before the [caprieval] module, the analysis will also be extended to the cluster level, providing a more robust analysis.

Analysis command line interface

The haddock3-analyse command line generates interactive plots from the data obtained by a [caprieval] module directory.

Here is the documentation of the haddock3-analyse

Dos and Don'ts

Don'tDo instead
take blindly the first ranked model/clusterconsider/examine multiple models/clusters, especially if they overlap within standard deviations in their score
consider the size of the cluster as an indication of its qualityuse the cluster score and not its size for selecting the best solutions (of course it is nice if the largest cluster is also the best scoring one)
consider/look only at the best model of a clusterwithin one cluster, do visualize and compare several models (e.g. the top4) to get an idea of the precision and make sure the clustering worked properly
take scores as proxies of binding affinity to compare different complexescompare scores only within the same system/complex (i.e. to distinguish models for one docking run), or run [prodigy] module

Any more questions about the analysis of the HADDOCK run?

Have a look at: