druglike filter

A druglike filter is a set of simple computational rules that weeds out molecules unlikely to make good oral drugs, the way a bouncer turns away guests who obviously do not meet the dress code before they reach the door. Applied to a huge library, it removes the most hopeless candidates fast and cheaply.

Most filters check easily computed properties against thresholds derived from known drugs: molecular weight, lipophilicity, numbers of hydrogen-bond donors and acceptors, polar surface area, and rotatable bonds. The best-known example is Lipinski's rule of five, a guideline for oral absorption. Other filters remove molecules containing reactive or assay-interfering substructures, such as PAINS or structural alerts, that tend to cause misleading results or toxicity.

The honest caveat is that these are soft guidelines, not laws. Many successful drugs break one or more rules, and whole modalities such as macrocycles deliberately live outside classic druglike space. Filters are best used to triage and prioritize, removing the clearly problematic while flagging, not banning, the merely unusual. Treating them as hard cutoffs can discard genuinely promising chemistry.