Learning biomolecular absorption spectra in graphene nanopores

Solid-state nanopores can be combined with optical measurements for the detection of biomolecules. With a view to a reliable optical biosensing library and avoiding computationally demanding simulations or expensive experiments, we develop a learning approach based on deep neural networks that are being trained on electronic, conformational, and optical characteristics from density functional theory (DFT) simulations of single amino acids placed in a graphene nanopore. Principal component analysis is used to compress the high-dimensional data into a latent representation, improving training efficiency while preserving the most important physical features. For the five amino acids considered in this work, the learning model predicts the full orientation-dependent absorption spectra with consistently high accuracy in previously unseen molecular orientations within each amino acid dataset, with test R ² values exceeding 0.9. These results indicate robust interpolation across rich conformational variations of the exact amino acids studied here. The developed workflow enables the construction of reliable optical biosensing libraries with near-DFT accuracy at a fraction of the computational cost. It thus provides a scalable and efficient pathway towards real-time, high-precision optical identification of biomolecules.