PAINS filter is well know molecular filter to remove promiscuous compounds for HTS.
Many computational tools implement PAINS filter as a drug like filter. I also use the filter but I do not confirm the reliability with our in house dataset.
Here is a report from researchers in Lilly about investigation of PAINS filter with their in-house dataset.
It’s worth to read I think. And also the author is also published about Lilly’s MedChem filter some years ago.
URL is below.
They used their compounds and assay data reported XC50. And single point data was excluded from the analysis. Also they omitted compound with less than 80% pure.
The analysis performed with several assay format, Alpha Screen, ELISA, FilterBinding, Fluorescence Polarization, FRET etc.
As interestingly only 2/6 assay formats showed more than 1.5 odd ratio of PAINS filter enrichment. The assay format is AS and FRET. This enrichment comes from very few number of alerts such as anil_di_alk_A.
They mentioned that High Hill slope (they defined >2.0 is high) compounds show high enrichment.
I need to care about what kind of assay format will use, and Hill slope of dose response to make right decision.