More physicochemical musings & ChEMBL

Following on from his earlier papers (e.g. ADMET rules of thumb for candidate drug design), Gleeson’s most recent publication in Nature Reviews|Drug Discovery details a now familiar tale of physicochemical woe associated with the failure of many medicinal chemistry programmes to focus adequately on properties other than target affinity.  Once again arguing that an over-reliance on targeting high potency in order to minimize the clinical dose and risk of toxicity related attrition (see now classic papers & citing works by Leeson and Hann) has continually seduced medicinal chemists away from ‘drug-like space’.  These are points which in themselves are worthy of further emphasis, but of particular interest in this case is the origin of the data in the ChEMBL database, a publicly accessible database of targets and drugs compiled by the European Bioinformatics Institute (EBI).  Whilst some of the available data are incomplete (and the authors acknowledge the limitations in their analyses) this database is an important and often under-utilized source of information and is worthy of wider attention.  In contrast to many of the physicochemical analyses published to date, all the ChemBL source information is freely accessible for others to validate and/or challenge.  The database contains over half a million compounds (stored as 2D structures), with accompanying in vitro biological data and calculated properties.

For this analysis, the authors calculated an ADMET score (i.e. the deviation from oral drug space as defined by AlogP and molecular mass) to characterize the data set.  The plot below shows 1791 oral drugs colour-coded according to their ADMET score (on the left), and then (on the right) the same drugs compared with the scores for the total content of the ChEMBL database (approximately 200k compounds).  A greater deviation from oral drug space is observed in the latter case (14% of drugs have ADMET scores >2 compared with 39% of ChEMBL molecules).

Further detailed graphics and analyses are also presented to arrive at the following conclusions:

  1. Average oral drug potency is approx 50 nM (Another clear observation was that 8% of oral drugs have both mol mass > 400 and AlogP > 4, compared to 30% of all ChEMBL molecules and 41% of ChEMBL molecules with nanomolar potency).
  2. The majority of oral drugs have off-target pharmacological activities (This analysis was clearly complicated by any intentional poly-pharmacology).   The graphic below shows, for 392 oral drugs, N = number of off-target hits with reported potencies ≤ 1 µM (i.e. only 29% in the N=0 segment are totally selective) :                                                                                    
  3. There was no clear relationship between in vitro potency and therapeutic dose (which is perhaps unsurprising given the numerous additional factors involved).

Whilst the points made in the paper mainly serve to reinforce the conclusions of earlier publications, they nonetheless highlight the importance, and usefulness, of open-source information sources and the prospect of their more widespread utility in future.

This entry was posted in Med Chem Strategy, Reviews, Software and tagged , , , , , , . Bookmark the permalink.

1 Response to More physicochemical musings & ChEMBL

  1. milkshake says:

    when discussing drug-likeness, it is problematic to use datasets that throw together drugs from different therapeutic areas, with differing mechanism of action. For example all beta lactam penicillin antibiotics are exceptionally ugly from the drugability standpoint (covalent binders, contain labile group and functional groups that are metabolically liable, they have lots of polar groups and mock Lipinski rules of thumb) and yet they work great because they were selected by the nature, and more importantly they do not have to be cell permeable to act as bactericide. On the other hand, CNS-active compound design needs to be very conservative (small, somewhat greasy) because lots of things will ruin BB permeability – but then there are classes of CNS drugs that get taken into brain by active transport and thus their properties are outliers when taken together with other CNS compounds.

    I also think it does not make good sense to have cell-surface receptor binders or drugs that target coagulation cascade in the analysis that throws them together with drugs that have to reach their target in cytoplasma. And then you have some drugs that have to be metabolically activated, or have a long-lived major active metabolite (which is accumulated or perhaps re-cycled through bile secretion).

    So these comparisons need to be made more meaningful by separating the datasets by the mechanism of action and the drug target organ. The best guide for the drug design is to look what has been done within a particular therapeutic area, and use examples of unrelated drugs with a healthy dose of skepticism. I understand why people seek to have universal rules but overly broad generalizations are actually harmful because they actually obscure the reasons for which drug candidates work or fail and they replace medchem insight and common sense with management-friendly dogmas.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s