Validations → CDK2 enrichment tests

• Method
• Enrichment Analysis
• Docking Results

Method

Two enrichment testing studies were performed. 

In the first, a collection of 57 unique CDK2 protein-ligand complexes available in the PDB were used in this study.  The protein ligand complex denoted by the PDB code 2R3J([1]) was selected as the base query structure. The SMILES strings representing each of the CDK2 ligands were extracted and passed into the MOE conformer generation engine to generate conformational models for each molecule.  The active set (the PDB ligands) together with approximately 900,000 drug-like molecules from the InhibOx CSpace database were filtered, aligned and docked using the DrugFinder platform.  The hits were ranked using the docking score as measured using the PLP scoring function. For comparison, a measure of shape similarity to the query ligand was also calculated for the hits, using the Tanimoto similarity index.

A second test of enrichment was conducted using the CDK2 ligands obtained from the DuD database ([2]). The actives and decoys obtained from this database were used to generate multiple-conformers and then passed through the DrugFinder service. The docked molecules were then inserted into the output database obtained through screening CSPACE and sorted according to scoring function values.

Enrichment analysis

Figure 1 shows the performance of the DrugFinder service as a Receiver-Operator Characteristic (ROC) curve, a standard analysis methodology.

Many of the CDK2 known actives downloaded from the PDB were analogues to the 2R3J query molecule, and therefore initial performance on the Tanimoto similarity measure is good. The percentage of actives found on the first 100 hits of the Tanimoto-based lists was approximately 56%, which was quite encouraging. The equivalent percentage for the PLP-based list was slightly lower at 45%. However, when the first 500 hits were considered, the PLP-based lists enrichment rate rises were significantly better than those of the shape based lists; around 91% on the PLP list, approximately 70% for the shape-based list.  

These enrichment studies have shown good performance when considering the fact that these hit lists were generated after scanning a database containing in excess of 900,000 molecules.

 

Docking Results

Figure (3) displays the docking poses of the two highest scoring hits obtained from screening CSPACE on DrugFinder against the CDK2 target. Both ligands show good complementarity of shape and functionality with the acive site. The known active (i.e. the 2R3J known active) is shown in red, whereas the two hits are coloured in blue.

 

References