Sequence signatures for the molecular differentiation of pnictogen- binding bicyclic peptides

The molecular differentiation of chemically similar, thiophilic heavy pnictogens remains a key challenge in metallopeptide engineering. Here, an integrated framework combining k-nearest neighbors (KNN) classification, residue positional mapping, graph-theoretic co-occurrence networks, and sequence physicochemical profiling was applied to over 12,000 Cys-constrained bicyclic peptides to decode arsenic(III) and bismuth(III) selectivity. The best model achieved robust minority-class discrimination on unaugmented test data (Recall=0.85, AUPRC=0.75). Based on the statistically validated sequence signatures, As(III)-binding motifs favor extended conformations, higher hydrophobicity, and elevated partial specific volume. In contrast, Bi(III)-selective motifs favor flexible, hydrophilic loops enriched with aromatic and imidazole residues. These sequence signatures provide granular, actionable rules for the de novo design of selective biosensors, scavengers, and chelators for heavy element discrimination, as well as for fostering a deeper understanding on how proteins discriminate metals.