The NIH has emphasized the need for benchmark datasets and robust method evaluation to improve reproducibility in computer-aided drug design (CADD). To this end, the Drug Design Data Resource (D3R) hosts community challenges using private datasets that are subsequently released into the public domain. These challenges focus on the prediction of protein-ligand complex structures and binding affinities etc.<b> </b>Our collaborators at the RCSB and wwPDB have recently changed the process by which new protein-ligand structures are released to the public. In the new process, a pre-release file containing protein amino acid sequences and their bound ligands is published five days prior to the public release of the corresponding 3D protein-ligand structures. Leveraging this process, D3R has developed Continuous Evaluation of Ligand Pose Prediction (CELPP), a weekly community challenge that evaluates whether CADD methods are able predict these 3D complexes before they are released.<br/><br/>A D3R server hosts these weekly challenges in three stages: Stage 1 runs when the PDB pre-release information becomes available on Friday night. CELPP downloads the pre-release protein sequences and ligand InChI strings from the PDB website, then runs a BLAST search of the sequences against the current version of the PDB. By screening the BLAST results using specific filtering criteria, a set of “candidate” proteins for pose prediction is selected. In stage 2, CELPP creates a “challenge package” containing these candidate proteins and the pre-release ligand InChIs, and sends it to all participants. Stage 3 runs on Tuesday, upon release of the new 3D PDB structures. CELPP superimposes the contestant-submitted protein-ligand complexes to the “answer” structures, and calculates the RMSD between the docked ligand and the crystal ligand. An evaluation report is sent to the participants. In this presentation, we will discuss the details of the filtering criteria in stage 1 and why we use those criteria to choose the docking candidates. Evaluation using other methods, such as RSCC and protein-ligand fingerprint will be compared to the traditional RMSD evaluation.

Selection of cross-docking candidates and comparison of evaluation metrics in Continuous Evaluation of Ligand Pose Prediction (CELPP)

253rd National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.