Macrocycle align
Brikard, Libmol , RDKIT, openbabel, Vina, libsampling
Assumed pH 7.4
Tautomers considered
Gasteiger charges
Proteins where prepared in CHARMM19 forcefield. Small molecules where parametrized using AnteChamber.
Iterated Local Search global optimizer search method
Exhaustiveness=50 #exhaustiveness of global search (default=8)
Vina scoring function (empirical + knowledge-based function)
For each ligand we have identified several best templates based on Tanimoto similarity . Further on we have generated free conformations of each molecule using Brikard for macrocycles (10,000 poses) and RDKIT for regular molecules. After that for each ligand and template we have identified matching points, and the free conformers were aligned to the template ligand. After that we have minimized the pose starting with harmonic constraints to reference points in presence of protein. Finally the molecules where freely minimized, with rotation of torsions not present in the template. Results where clustered. In addition for the molecules containing rotational bonds, not present in the template, we have run local Monte Carlo Minimization. For scoring we have used the following measures. 1) Vina energy score; 2) RMSD to the template after free relaxation (stability) and finally size of the cluster (as same ligand template pair may have several alignments). For each cluster we have chosen lower RMSD representative, and sorted the largest cluster based on Vina score.
Yes
GradientBoostingRegressor
RDkit, sklearn
learning rate 0.1
n estimators 300
subsample 0.5
We used RDkit package to convert SMILES strings into vectors of fingerprints + chemical descriptors.
From ChemBL database we fetched 2873 ligands most similar by cosine distance to the proposed targets. This
set was used to train GradientBoostingRegressor provided with sklearn package.
No
Yes