GPCRs structural biology 2

Question 1

Timeline of drug discovery

Answer 1

Target selection, hit finding (using HTS, in silico docking, DNA encoded, etc…), SAR/hit optimisation, lead optimisation (in vivo), and clinical trials.

Question 2

B-AR overview and drug discovery

Answer 2

First targeted GPCR.

The B2-AR has 2000+ ligands that act at it. The beta-hydroxy group is what makes the endogenous ligands catecholamines

Question 3

How machine learning is developed

Answer 3

Using HTS data, chemical groups, and structural shapes will be identified for having activity at the receptor. Algorithms learning the behaviours of drug-receptor interactions can predict the effects of different chemicals at the receptors, and identify ideal shapes and functional groups for drug discovery.

It will often utilise public datasets, such as ChEMBL for a high number of data sets to base its algorithm off of.

It cannot be very accurate due to the dynamic nature of receptors in vivo. The database structures are static. As such, kinetics and binding cannot be entirely accurate.

Furthermore, in silico docking modelling will create many (90%) wrong binding poses so the datasets must be filtered.

Question 4

What are AIFs and how are they used

Answer 4

They are atomic interaction fingerprints.

Describe ligand-receptor interactions on an atomic level. Chemical and physical interactions. Can be used to predict drug binding.

It utilises databases to identify specific interactions in the binding site with the ligand to identify the interactions that are specific for agonism/antagonism. By identifying the functional groups and atoms that drive the binding and pharmacological mode of action, it can be utilised in drug discovery.

Question 5

Machine learning software example and its uses

Answer 5

XGBoost

It extracts detailed SAR from the full dataset (the noisier dataset/non-filtered).

Not only used to identify the specific interactions of the drug with the receptor, but to also identify what residues/atoms are driving the drug’s selectivity for specific receptor isoforms.

It identifies the specific pharmacological roles for atomic interactions between the receptor and the ligand.

Being able to predict these pharmacological properties is dependent on the quality of the docking

Question 6

Future of structural biology

Answer 6

Potential for identifying universal “rules” that the receptors follow.

E.g., tryptophan residue is conserved in the binding site of GPCRs. Interactions with this tryptophan prevent the conformational change that mediates G-protein signalling. Potential universal target for antagonists.

Other potential uses of machine learning in application of this knowledge to diseases.