Visualizing Antipsychotic Receptor Affinity : Part One
Introduction
For a while, I've been looking for a resource that collects together data on receptor binding/affinities and different classes of medications. As an example, say I want to compare the 5HT1A and D2 affinities of two different medications, I'd have to try and find either some literature or use an internet search to find the data (years ago, I tried manually collating this data, but it got boring very quickly).
Bryan Roth and Estela Lopez at the University of North Carolina, Chapel-Hill host a fantastic public domain database: the PDSP Ki database [1] which contains over 64K ligand-receptor pairs and their Ki values (in nano-mole units), which can be downloaded as CSVs for data mining.
Data Mining and Preprocessing
The complete PDSP database contains ligand-receptor data on multiple species (e.g. human, pig, mouse etc.) and tissue types (central / peripheral nervous, vascular etc.). I filtered the data as follows:
- Select only "HUMAN" species and subset on data for ligand-receptor combinations studied in CNS tissue or cloned receptors
- Filter on 20 common antipsychotic medications from [2]
- The data is trimmed to remove receptors for which there is very little data (< 5 receptor Ki values) Remove medications for which there are no records for any receptor.
- Where there are multiple Ki values for the same receptor-ligand pair, similarly to the PDSP Ki database's query interface, we compute the dispersion, discard values at the tails of the distribution and then report the resulting average. The PDSP interface use the mean, and then +/- 2SD, but I used the median and then +/-1.5 IQR (as samples sizes are small and there's no reason to assume the distribution is symmetric).
Processing for Visualization
Next, we want to visualize the receptor-ligand affinities and each is described by the inhibitory constant Ki, for which large values represent low affinity and vice versa. They range across orders of magnitude, so we instead use the pKi, defined as the negative log (base 10) of the Ki value e.g. pKi = -log10( Ki ). As pKi is on the logarithmic scale, a one unit change in pKi represents an order of magnitude change in Ki. Medications with higher pKi (toward or greater than 9) having higher affinities and vice versa.
Interpretation
This results in the heatmap diagram shown above, generated using R and the plotly API to make it interactive. It's best viewed on a larger browser window (click here), where hovering over a 'cell' in the grid shows the receptor, ligand and numerical pKi value, and the interactive zoom functions work way better.
Bare in mind that we're looking at an average of samples for each receptor-ligand pair (collated from the PDSP database) and some of these samples have wide distributions, so the numerical values may not marry perfectly with specific values published elsewhere. Some medications have relatively few receptor studies, and where data is not available, the cell is filled white. Also, I have not aggregated together differently-named entries in the PDSP database e.g. "DOPAMINE D2" with "D2" - because at this stage, we just want to visualize the data (see Next Steps below).
Next Steps ...
Visualizing this stuff is interesting, but is essentially just a graphical table of values. In the next post, I'll look at ways of visualizing the data that emphasizes comparison of the binding profiles of different medications. I'll also include links to a repository with the code as it develops, so you can interrogate the PDSP database for any other medications/receptors of interest.
References:
- The Multiplicity of Serotonin Receptors: Uselessly diverse molecules or an embarrasment of riches? BL Roth, WK Kroeze, S Patel and E Lopez: The Neuroscientist, 6:252-262, 2000
Taylor, D., Paton, C., & Kapur, S. (2015). The Maudsley Prescribing Guidelines in Psychiatry. 12th Edition, Wiley-Blackwell.
