Highly efficient chiral discrimination of proteinogenic amino acids combining machine learning with terahertz time-domain spectroscopy

Chiral molecular discrimination is critical for drug safety and disease diagnosis, yet discriminating amino acid enantiomers remains challenging due to the limitations of chiral selectors. Terahertz spectroscopy captures molecular structural vibrations with low photon energy, but conventional methods fail to resolve subtle chiral differences in biological systems. Here, we integrate terahertz time-domain spectroscopy with a machine learning framework based on multi-feature input to achieve high-precision chiral discrimination of enantiomers across proteinogenic amino acids. We constructed a comprehensive database of 11,700 terahertz spectral profiles and trained six machine learning models that demonstrated exceptional performance: over 97% accuracy in chiral discrimination and 99% in classification. The method establishes a rapid, cost-effective platform for chiral analysis based on terahertz technology, advancing its applications in pharmaceutical quality control and biomedical diagnostics.