A Canonical Text Representation for Polymers via BigSMILES and Tree Automata

A BigSMILES string encodes the structural connectivity of any polymer chemistry and topology as a linear string. However, multiple BigSMILES strings can encode the same ensemble, making string-based searches for polymers in digital databases challenging. This work presents a canonicalization algorithm that breaks the degeneracy of the BigSMILES language for both linear and branched polymers and can reverse-translate canonicalized structures back into BigSMILES. The algorithm was validated on broadly representative polymer chemistries and topologies from the literature. First, the BigSMILES string is mapped onto a tree automaton, a type of state machine that accommodates branch points and recognizes the same ensemble of molecules that BigSMILES encodes. The automaton can then be minimized into a unique graph with the fewest states through existing algorithms. Finally, a human-readable canonicalized BigSMILES is obtained upon translation of the state machine transition rules back into a string. This robust canonicalization algorithm allows polymers to be searched rapidly in large database systems, making data findable, accessible, interoperable, and reusable (FAIR) and enabling the development of novel data-driven approaches with BigSMILES.