Comparing Molecular Patterns Using the Example of SMARTS: Theory and Algorithms

Molecular
patterns are widely used for compound filtering in molecular
design endeavors. They describe structural properties that are connected
with unwanted physical or chemical properties like reactivity or toxicity.
With filter sets comprising hundreds of structural filters, an analytic
approach to compare those patterns is needed. Here we present a novel
approach to solve the generic pattern comparison problem. We introduce
chemically inspired fingerprints for pattern nodes and edges to derive
an easy-to-compare pattern representation. On two annotated pattern
graphs we apply a maximum common subgraph algorithm enabling the calculation
of pattern inclusion and similarity. The resulting algorithm can be
used in many different ways. We can automatically derive pattern hierarchies
or search in large pattern collections for more general or more specific
patterns. To the best of our knowledge, the presented algorithm is
the first of its kind enabling these types of chemical pattern analytics.
Our new tool named SMARTScompare is an implementation of the approach
for the SMARTS language, which is the quasi-standard for structural
filters. We demonstrate the capabilities of SMARTScompare on a large
collection of SMARTS patterns from real applications.