A recent paper in JCIM caught my eye “Modeling hERG Channel Liability: From Structural Insight to Highly Accurate Qualitative and Quantitative Models” DOI . The paper includes a discussion of the binding site on the ion channel commenting on “The highly adaptive nature of the hERG ligand-binding site may poses challenges for structure-based approaches, such as molecular docking” together with a regression model and a classification model. The investigation of the binding site highlights a protonated nitrogen and key aromatic interactions.
They used a final descriptor set comprised 221 atom types and 40 correction factors, capturing whole-molecule features like molecular flexibility, fraction of sp2-hybridized atoms in a molecule.
They also provided the curated 8000 compound data set as part of the supplementary information. I had a quick look at the 8000 compound data set.
I imported the molecules into Vortex and calculated a variety of physicochemical properties and looked up development statues and the number of clinical trials reported (from clinical trials.gov) using a couple of Vortex scripts..
Looking at calculated physicochemical first, as might be predicted basic molecules tend to be more active at hERG, with acids and Zwitterions much less so. There is also some evidence that the more lipophilic molecules are more active.
Generating a TSNE plot[ https://en.wikipedia.org/wiki/T-distributed_stochastic_neighbor_embedding] and coloured by hERG class (0 less than 10000 nM, 1 is greater than 10000 nM) used in the publication. Firstly it is clear that hERG activity is widely distributed within the chemical space encompassed in the dataset. It is also a nice way to spot where small changes result in a significant modulation of hERG activity.
Looking at just the compounds that clinical development has been reported it is clear many have reported hERG activity below 10 uM.
Whilst most of the compounds are in PubChem, 4772 are also identified in patents, 498 are ligands in the PDB (but not necessarily for hERG).
