BigFingerprint: A Topology-aware Extended-connectivity Fingerprint for Polymers

Machine learning-driven polymer design and property prediction require molecular representations that encode not only local chemistry but also the statistically distributed monomer sequences and connectivity patterns and global topology that distinguish polymers from small molecules. Existing fingerprinting workflows often fail to capture global polymer architecture, as well as connectivity across repeat units, end groups, and linking groups. This work develops BigFingerprint, a topology-aware extended-connectivity fingerprint for polymers represented by BigSMILES. BigFingerprint consists of two components, a polymer extended-connectivity fingerprint, p-ECFP, that encodes atom-centered substructures within and across stochastic molecular fragments, and a polymer topology fingerprint, p-TFP, that encodes global architectural features such as block, statistical, graft, star, branched, and network topologies. Through representative structural examples and polymer property prediction tasks, BigFingerprint retains the chemical interpretability of circular fingerprints while extending their applicability to complex polymer architectures. As a critical component of the BigSMILES-centered ecosystem, BigFingerprint integrates local substructural information with global topological descriptors to provide a general and chemically interpretable representation for substructure analysis, similarity