5zcmb-LigandScoringProtocol.txt

Name

Rhodium HTS

Software

Pymol 1.4.1, OpenBabel 2.3.90, Rhodium 380E7

Parameters

Docking resolution 1.7
Refined Poses: 144
Trial Rotations in SO3 rotation group, per conformer, per Grid Point: 72

Method

Rhodium docking was performed with default "HTS" settings. An automated grid search was peformed over
the entire strucuture of the protein model. A low resolution score for each ligand was computed at each point in the grid using a step function
similar to the of Vakser, et. al. Multiple fixed conformers were considered, as well as additional ligand models created though a
geometic reflection operation, namely by inverting the sign of x-coordinate of each atom.
This was a primitive approximation for considering stereoisomers.
Geometric coordinates of the higest-ranking poses (by the low resolution method) were optimized in their location by minimizing
a docking score with respect to coordinates.
The geometery optimization method was the Multidirectional Search of V. Torczon, using seven coordinates of each ligand, defined by the
center of mass (3-vector) and a quaternion (4-vector) representing the orientation. The minimized function was a docking score
based roughly on the published Autodock Vina score but modified by subsitution of highly-accelerated exponential function, for the
standard C exp() function, and taking advantage of SIMD instructions on a hardware level. The weight of one AutoDock Vina term was modified.

Final pose ranking was perfomed not with the Vina-like score, but with
a score called KRh, which includes factors such as unsatisfied hydrogen bonds, cavity-filling, molecular weight, and
a molecular-weight corrected docking score, the latter crudely representing the enthalpic efficiency.
The weights of these factors were previously determined by linear regression for predicting IC50s of
protein-ligand complexes in the BindingMOAD database.

5zcmb-PosePredictionProtocol.txt

Name

Rhodium HTS

Software

Rhodium 380E7, OpenBabel 2.3.90, Pymol 1.4.1

System Preparation Parameters

Geometry optimization with obminimize -n 20000 -ff MMFF94,
Conformer generation with confab - obabel --confab --conf 90000000 --ecutoff 4

System Preparation Method

Ligands were prepared using OpenBabel to minimize the optimize the free ligand geometry.
We used the MMFF94 forcefield. Multiple conformers were generated with the Confab program available in OpenBabel.
Due to the count of rotatable bonds being relatively high, the number of allowed conformers to be sampled was simply set to a very high number,
90 million. However an energy limit of 4 (4 kcal/mol?) was applied to filter out strained geometries.
Rhodium uses an all-atom (hydrogen included) docking method, so hydrogen atoms were added to each protein model using Pymol 1.4.1.

Pose Prediction Parameters

Resolution of Search Grid=1.7 Number of trial orientations at each grid point=72 Poses optimized following the low-resolution search=144

Pose Prediction Method

This is a high-thoughput, fixed conformer docking method.
Rhodium docking was peformed using a set of fixed parameters defined as our "HTS" protocol.
These simply define the resolution of the search grid (1.7 A), number of trial orientations at each grid point (72),number of iterations
perfomed in the MDS (Multidirectional Search) algorithm, and number of poses to be optimized following the low-resolution search (144)
Docking was peformed in a fully automated mode, with no additional parameterization.
Ligands were prepared starting from geometries provided (S-steriosiomer, if applicable).
During docking, mirror images of the ligands were considered, in an primitive, high-speed fashion. These mirror images are formally the
R-steroisomers if no additional sterocenters are present.