Name

LIE with bootstrapping

Software

OpenBABEL-2.3.2
AutoDock Vina-1.1.2
GROMACS-4.5.5
y_tb-0.2.4 (in house)
various small in-house utilities

Parameters

please see supplementary for GROMACS mdps
dodecahedron shaped periodic box with ~15K TIP3 explicit waters and 0.1M NaCl at pH=7.0
AMBER-ILDN parameters set for protein atoms
adjusted YFF1 with Kirchhoff charges for compounds; bonded part from GAFF

Method

Note. It's 'set1', this set is made with bootstrapping of free energy subsets to parameterize more adequate LIE model.
A linear interaction energies (LIE) model (one per pose) consists of 8 replications (from the same starting coordinates but different random seed in velocities generation routine), each of 4 ns equlibrum MD run followed by 2 ns of productive run with statistics being collected. In contrast to set0 here we added estimations from free energy runs over 18 compound of FEP set 2 to the subset of 12 reported in literature compounds in order to parameterize the 7-parameters weighted LIE model: dG=a*Q_drug+b*LJ_drug+g*Q_site+d*LJ_site+e*Q_drug_site+n*LJ_drug_size, {the fit LIE model is a=-0.078139, b=0.147434, g=0.012377, d=0.075575, e=-0.018091, n=-0.152529, Q_site_0=+6418.26KJ/mol, LJ_site_0=-1458.75KJ/mol }, for weights calculations dG was also multiplied by factor 0.02 which doesn't have good theoretical justification but known to work considerably better in practice [our experience plus for example ref Almlof et al, 2007].
The most important difference from stage 1 of the competition is bootstrapping. In addition to 12 literature compounds there were 18 compounds with energy measured precisely by free energy technique (set#2 of free energy part of the competition). See my corresponding free energy report about the technical details. Alltogether 30 items were used to parameterize LIE model. With a bigger data set, an extended 6-parameters LIE model was used. The LIE energies were converted into kcal/mol and stored as csv file in this archive.

Name

AI_MD

Software

AutoDock Vina-1.1.2
various small in-house utilities

System Preparation Parameters

please see supplementary for Autodock vina.cfgs

System Preparation Method

Compounds were prepared previously at stage 1 of the challenge.
2 receptor structures were taken from resolved pdbs (FYMF and KJYP), non-protein molecules are cut and 24 x 24 x 24 A binding site was defined manually. The protein was threat as rigid during docking stage.

Pose Prediction Parameters

For each of the 2 selected conformations of FXR protein up to 20 poses were generated with Autodock Vina (see example of autodock parater file in the SuppInfo).

Pose Prediction Method

We simply docked each compound into manually defined binding site and selected the poses that have the smallest RMSD of their 3D voxel maps with a structure from pool of stage 1 resolved structures.