Single molecule identification of d- and l-amino acid enantiomers is highly desirable but challenging due to subtle chemical differences, dynamic interconversion, and the coexistence of numerous post-translational modifications (PTMs). Notably, proteomics has not yet achieved the level of precision seen in genomics and transcriptomics, largely because existing methods struggle to resolve chirality-specific molecular signatures within complex proteomes. Here, we combine a data-driven machine learning approach with quantum tunneling to screen and detect d/l-isomers and PTMs, including isobaric isomers, methylation, phosphorylation, and ring formation. Our ML framework should be distinguished from ML-for-materials paradigms focused on replacing expensive quantum calculations or learning transferable descriptors. Instead, it serves as a decision-theoretic layer that maps high-dimensional, precomputed quantum tunneling features to biochemical labels, addressing the analytical challenge of signal disambiguation rather than computational acceleration. Our work illustrates a systematic and internally consistent physics-based simulation framework for identification of amino acid d/l-isomers and variants and thus opens frontiers in challenging single-molecule protein sequencing.
Mittal et al. (Wed,) studied this question.