Selectivity mining – multiple activities in Activity Miner
- 3. © Cresset
Activity Cliffs
> Many names:
> Disparity (Merck 1990s)
> SALI (Guha/Drie 2008)
> Activity Landscapes
> Activity Cliffs
> For each pair of molecules
𝐴𝐴𝑐𝑐𝑐𝑐1 − 𝐴𝐴𝑐𝑐𝑐𝑐2
𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎12
> Usually distance = 1 – similarity
> Similarity in 2D
3
- 4. © Cresset
Where 2D similarity causes problems
> Bioisosteres – low 2D similarity but biologically
similar
> Enantiomers and Chirality
> Locality issues
Fingerprint locality (ECFP4 sim=1.0)
N
N
O
Cl
O
Cl N
N
N
O
O
N
O
O
O
O
N
N
NN
4
- 5. © Cresset
Using 3D molecular similarity
> 2D metrics are easy: 1:1 map to topology
> 3D is defined for conformers, not for molecules
0.66
0.92
5
- 6. © Cresset
Context required for 3D similarity
> Need context for 3D similarity – bound to the
protein
> Align all molecules in the active site
> Ligand view of alignment more informative than a
protein view?
N
N
S
N
O
O
N
N
S
N
O
O
N
N
S
N
O
O
N
N
S
N
O
O
N
N
S
N
O
O
Protein ‘view’
of alignment
Ligand ‘view’
of alignment
Signal
N
O
N
6
- 7. © Cresset
3D disparity in Forge’s Activity Miner
1. Generate conformers
2. Align to reference(s)
3. Calculate similarity matrix on aligned
conformations
> Allow small movements
4. Calculate disparity matrix from similarity
numbers
> Similarity cutoff of 0.95 (Distance cutoff of 0.05)
5. Visualise
> Difficult – 100 molecules gives 4950 pairs!
𝐴𝐴𝑐𝑐𝑐𝑐1 − 𝐴𝐴𝑐𝑐𝑐𝑐2
𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎12
- 10. © Cresset
Activity View
Ten Nearest
Compounds,
height = distance
Shade = Disparity
Strong colours = Strong SAR
Current Focus
Compound
Comparator compound
10
- 13. © Cresset
Detailed Electrostatics decode SAR
Has an effect here
Also has an effect here!
Difference plot – Regions where each molecule has stronger electrostatics
13
- 14. © Cresset
Selectivity Cliffs
> Selectivity often as important as potency
> Look at what structural changes caused large
changes in selectivity
> Now our visualisation problems are even worse!
𝜅𝜅 ≈
∆ 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆
∆ 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆
=
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝛽𝛽
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑎𝑎
𝐴𝐴
−
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝛽𝛽
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑎𝑎
𝐵𝐵
1 − 𝑆𝑆𝑆𝑆 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆
14
- 19. © Cresset
More Electrostatic Differences
> Explain changes in selectivity (how does an
indazole compare to a 2-aminopyrimidine?)
α/β=8.05/8.3 α/β=7.3/5.9
19
- 20. © Cresset
Conclusions
> Activity Cliff/Disparity analysis provides quick insights
into SAR
> Focus on understanding the reason for a cliff
> Drive design decisions
> Multiple ways to navigate the data
> Compound focus > Most significant changes
> Global overview > Cluster analysis
> 2D and 3D both useful
> 2D provides insights into conformational changes
> 3D provides insights into electrostatic effects
> Visualising multiple activities simultaneously allows
selectivity analysis
20
- 21. © Cresset
Acknowledgements
> Andy Vinter
> Founder Cresset
> Tim Cheeseright
> Director of products
> Rae Lawrence
> North America sales and support
> Nigel Palmer
> Developer
21
- 22. © Cresset
> 3D Activity Cliffs available in Activity Miner module for
Forge and Torch
3D Activity Cliffs
3D Design tool, SAR interpretation
SAR interpretation & Activity Cliffs, 3D Design,
3D QSAR, Pharmacophore modeling
22
