Current methods of ligand fitting that are based on either topological analysis of electron density [Menéndez-Velázquez-2003] global optimization of position and conformation of a ligand in a density blob [Diller-1999] interatomic distance matrix [Koch-1974] [Cascarano-1991] or on varying torsion dihedral angles of shape-matched ligand conformations [Oldfield-2001] are unable to prevent creation of high energy, sometimes even chemically unrealistic, ligand models. As a result, there are a number of PDB ligands with unlikely, very high-energy structures [Perola-2004]. For example, the PDB structure of an inhibitor of RNA polymerase in 1nhu has significant repulsion between the two methylene groups.

FLYNN is composed of two main components, location of ligand density and ligand fitting. Location of density generally works well with clean density but there are times when ligand density is unclear or not well resolved. In the latter cases, the user should supply a bounding box or simply just input the ligand density by itself and use the density as is.

By default, FLYNN samples bioactive conformations [Boström-2001] [Boström-2003] of the input ligand, however, there are times when it might be desirable to input the ligand and use it as is (see section Required Parameters for more details).

Once the location of the ligand in the map and the conformations have been selected, FLYNN then adapts the initial conformations to the ligand density using a modern force-field, MMFF94 [Halgren-I-1996] [Halgren-II-1996] [Halgren-III-1996] [Halgren-IV-1996] [Halgren-V-1996] [Halgren-VI-1999] [Halgren-VII-1999]

The potential function being used to adapt the ligand is

\(V = V_{ff} + \lambda V_{shape}\)

where \(V_{ff}\) represents the internal energy of the ligand and \(V_{shape}\) is the overlap between the ligand and the electron density. \(\lambda\) is a mixing parameter that represents the degree to which the shape of the density dominates the combined potential during the current optimization step [Wlodek-2006].

The strain placed on the ligand is bounded while the function is optimized producing high-quality fits with low-strain ligand conformations.

Fragment Fitting

Fragments are fit taking the input fragment cocktail and, one at a time, fitting each fragment against each region of detected density. Once a fragment has been placed, it is further analyzed to ensure that all possible orientations of the fragment have been sampled. In poor density, several orientations may fit equally well. To break ties, FLYNN scores each pose with the following scores:

o RSCC (real space correlation coefficient) This is a measure of fit to electron density (higher is better). [Jones-1991]

o PLP Piecewise-linear potential (lower is better). [Verkhivker-2000]

o Chemscore (lower is better). [Eldridge-1997]

The docking scores are not used to fit the molecule, they are only used to rank the output. Unless highly symmetric molecules are being input, the real space correlation coefficient (RSCC) is the preferred method of ranking results to density.

To use fragment mode, please add the “-fragment” option to the command line. In future versions of FLYNN, this will most likely become the default setting.

The output of the fragment fitting process is a file for each density region that includes the fragments fit to the region sorted from best-fit to worst fit. For example:

prompt> flynn -in fragments.smi -out 2IKO_cocktail.sdf ...

would result in the files:


one for each blob found. By default, the first molecule in the file is the best fit to the density (RSCC):

\(RSCC = \frac{\Sigma |p_{obs} - <p_{obs}>| \Sigma |p_{calc} - <p_{calc}>|}{(\Sigma |p_{obs} - <p_{obs}>|^2 \Sigma |p_{calc} - <p_{calc}>|^2)^\frac{1}{2}}\) [Jones-1991]

where obs refers to the experimental electron density and calc refers to the calculated electron density sampled from grid points around the residue being scored. The experimental density is usually the sigma weighted difference map when loading an MTZ map. The calculated density is generated by simulating scattering of the ligand conformation with bfactors set to 20.

To sort using another measure, for instance, PLP, use the sort flag:

prompt> flynn -sortBy plp -in fragments.smi -out 2IKO_cocktail.sdf ...


The -sortBy flag can be used in non-fragment mode as well.