Discrete Geometry Chemistry Chemical Shift Predictor: Ca Chemical Shifts in Small Peptides

Predicting NMR Cα chemical shifts from peptide structures typically requires computationally expensive quantum mechanical calculations or extensive empirical databases. The discrete geometry chemistry (DGC) paradigm offers an alternative: molecular properties approximated through geometric abstraction and minimal parameterization. This work demonstrates that Cα shifts in small peptides (6–60 residues) can be predicted using only backbone Cα coordinates and residue physicochemical properties. A linear ridge regression model with 27 features—12 geometric descriptors (k-nearest neighbor distances, radius of gyration, local density) and 15 physicochemical parameters (hydrophobicity, volume, secondary structure propensities)—achieves mean absolute error of 2.65 ppm across 585 residues from 21 NMR structures. The model generalizes well to ordered peptides (MAE = 1.92 ppm) but fails on proline/glycine-rich or disordered structures (MAE = 4.03 ppm). With microsecond-scale inference requiring only arithmetic operations, this zero-cost framework addresses 36,652 Protein Data Bank peptides lacking experimental Cα shift data, enabling rapid screening for peptide design and NMR assignment validation.