Regad et al. dataset

<strong>Description of the file DataDescription.csv</strong> <br>This study is based on a dataset composed of 483 pocket-ligand pairs. To create a large training set of pocket-ligand pairs, we initially gathered the refined set from the PDBbind database (Wang et al., 2004; 2005) and the Astex test set (Hartshorn et al., 2007) and selected complexes that contained drug-like ligands (i.e. small chemical compounds). Two protein-ligand complex datasets were compiled for the training set. The first one is composed of 560 non-redundant protein-ligand structures, with a resolution better than 2.5 Å retrieved from the refined set of the PDBbind database. From these structures, we removed those containing metal ions or cofactors next to the co-crystallized ligand resulting in a selection of 432 structures. The second one is composed of 85 manually curated protein-ligand complexes from the Astex test set. As for the previous set, we removed the structures with some ions or cofactors next to the ligand, resulting in a selection of 51 structures. The resulting dataset corresponds at the end to 483 protein-ligand structures. The id and information about each pair are available in the following table DataDescription.csv<br>This table contains: * PDB code of the complex * Protein chains containing in the PDB files. If the complex contains several protein chains, their chain Id are separated by "/" * Amino acid (AA) sequence of each chain. AA sequences correspond to sequence of the crystallized proteins. If the complex contains several protein chains, their AA sequences are separated by "/" * UniProt Id of each chain. If the complex contains several protein chains, their UniProt Id are separated by "/". NoId means no UniProt Id was find for a given chainSmile code of the interested ligandPDB code of the interested ligandObtained cluster of the pocket-ligand pairs