Cyclic Systems Distribution Along Similarity Measures: Insights for an Application to Activity Landscape Modeling

Abstract

The emerging concept of the activity landscape has been widely applied for structureactivity relationships (SAR) characterization. Since chemical space representation plays a crucial role in activity landscape modeling, an adequate selection of similarity measures is desirable. Herein a set of 658 cyclooxygenase inhibitors were structurally analyzed using 12 molecular similarity representations and two levels of chemotype classification. Then, three uncorrelated similarity measures and mean similarity (obtained with data fusion) were combined with chemotype information using the herein proposed chemotypesimilarity graphs. Chemotype-similarity graphs showed the cumulative distribution of molecular pairs with the same and different chemotype along similarity values; leading to establish an interpretable, quantitative and activity independent threshold in similarity measures based on chemotype distributions. This approach gave additional information to similarity measures and can be considered as an interpretable criterion to define high and low similar compounds. The results were applied to model the activity landscape using StructureActivity Similarity (SAS)-like maps. Some examples of pairs in each region of the SAS-like maps were analyzed which showed valuable SAR information for cyclooxygenase inhibitors.

Analyzing Multitarget Activity Landscapes Using Protein–Ligand Interaction Fingerprints: Interaction Cliffs

“…In this work, a threshold of one log 10 unit of absolute difference was used to distinguish between compounds with high and low potency. However, assigning a threshold for structural similarity is not straightforward as different criteria to impose thresholds can be employed. , In this paper, the thresholds for structural and interaction similarity were set as the mean similarity value plus two standard deviations and were calculated individually for each data set. A graphical representation of the density distribution for each data set is shown in Figure .…”

“…Figure 3A shows that in the complex 2WMR the pyrimidinone moiety of 1 forms a hydrogen bond with the E85 hinge.46 and C87 hinge. 48 amino acids located in the hinge area of CHK1 and also has hydrophobic contact with the gatekeeper amino acid L84 GK.45 (Figure 3D). 50 The same hydrogen bonds are present in the complex 2X8E by the triazolone of 2, but in this complex the 4-pyridyl ring is located in the solvent channel of the kinase domain forming hydrophobic interactions with S88 linker.49 and G89 linker.…”

“…In both crystal structures imatinib forms hydrogen bonds between the pyrimidine and the main chain of M hinge. 48 , the NH linker with the T GK. 45 , and the amide moiety of 3 with the side and main chains of E αC.24 and D xDFG.81 , respectively.…”

“…46 As discussed before, 49 chemical space and, hence, activity landscape models are highly dependent from the chemical representation used to describe molecules. In contrast with the strategy used in previous work where the combination of different molecular representations was needed (consensus activity landscapes), 7,19,48,49 here, only the 3D similarity (i.e., the ComboScore, which combines shape and pharmacophore similarity) was used due to the availability of the bioactive 3D conformation of the ligands. Although 3D techniques other than ROCS could have been used in addition as well.…”

Progress in the Visualization and Mining of Chemical and Target Spaces

“…SmAS maps are an extension of the Structure‐Activity Similarity (SAS) maps initially developed to explore the SAR for single targets 53,54. An application of SAS maps to analyze the SAR of cyclooxygenase inhibitors was published recently 55. This visualization tool is also related to the Dual and Triple Activity Difference (DAD/TAD) maps that were recently proposed to characterize the SAR of compound data sets screened against two or three targets, respectively 37,38.…”