An old colleague of mine Bob Sheridan has just published an interesting (and pretty comprehensive) paper in JCIM, Debunking the Idea that Ligand Efficiency Indices are Superior to pIC50 as QSAR Activties.

There seems to be a proliferation of various ligand efficiency (LE) metrics intended to give the scientist an objective measure of how efficiently a molecule is binding to the intended target Ligand efficiency: a useful metric for lead selection. Whilst I’ve always found them useful to get a feel for comparing the influence of substituent changes in QSAR studies, for example introduction of a large lipophilic substituent may buy affinity but at the expense of LogP. I do wonder at times if the LE numbers are assuming greater importance than the underlying biological data.

Recently several papers have use LE instead of pIC50 in QSAR models and claimed improvements, this seems a little counterintuitive and Bob Sheridan very nicely offers an explanation.

The apparent superiority is a statistical artefact and the improvement occurs when.

1. Ligand efficiency indices are highly correlated with the physical property included in their definition (number of non-hydrogens, ALOGP, TPSA, etc.),

2. The property is easier to predict than the original pIC50.