Activity cliff is very popular concept in medicinal chemistry. It means that similar compounds show very different biological activities. There many approaches to apply the concept to drug design.
BTW, what does ‘Similar compounds’ mean? In the area of chemoinformatics 2D fingerprint based similarity is widely used, such as Tanimoto similarity of ECFP, MACCS key etc. Sometime ECFP work well to picking similar compounds in 2D world also it works well not only compound clustering but also QSAR modeling.
QSAR is based on concept which similar compounds should have similar activities I think. So activity cliff is difficult to predict with QSAR model I think. Any way, ‘Similar’ is keyword for informatics.
Today I read an article about 3D based activity cliffs. The URL is below.
The author compared same molecule pairs with 2D similarity (ECFP-Tanimoto) and 3D similarity (cresset’s XedeX uses Shape and electrostatic feature).
Fig3. shows relationship between 2D similarity and 3D similarity and it indicates that many compounds show high 3D similarity even if 2D similarity is low. So 3D based similarity gives new insight of activity cliffs.
The Fig 4-10 compares 2D/3D molecular similarities. XedeX 3D similarity seems sensitive to stereo chemistry, substructure of hydrophobic groups, but not sensitive to bond order (Fig 6). So XedeX 3D feature seems more abstracted properties of molecules for me.
Their approach calculate 3D similarity and visualize compound pair with electrostatic nature. It helps chemist to understand 3D SAR for next compound design.
I enjoyed and learned many thing from the article but still thinking about are there activity cliffs in the real world? Because compound similarity is human defined parameter it’s variable, it depends on what kind of metrics do we use.
So cliff is not true cliff but sometime useful for detecting small change of structure which improves activity I think.