AI-Driven Two-Dimensional Molecular Electronics Spectroscopy for Chiral Discrimination of Natural Products

For centuries, natural products (NPs) have captivated attention with their rich stereochemical diversity and potent biological activities, serving as a cornerstone in drug discovery, particularly for cancer and infectious diseases. Here, we report a physics-informed ‘artificial intelligence-driven two-dimensional molecular electronics spectroscopy’ (AI-2DMES) framework that enables the chiral discrimination of isomeric NPs at atomic resolution. The method succeeds in the sensitive and specific identification of three enantiomeric, one diastereomeric, and two constitutional isomer pairs via pre-computed transmission and conductance signals. To mimic experimental conductance variability, we introduce conductance-distribution features derived from density functional parameterized tight-binding molecular dynamics (DFTB-MD), incorporating thermal fluctuations, multi-orientation molecular contributions, and signal broadening into the AI training framework. Logistic regression achieves 100% binary classification accuracy across all six isomeric pairs and 98% multi-class accuracy across 10 NPs. These results demonstrate that AI-2DMES encodes sufficient stereochemical information for isomer differentiation and could be a powerful and general platform for chiral sensing.