AI Reconstructs Molecular Structures From Simulated TERS Images
The SMARTERS model localized atoms with sub-0.1 Å error in simulated data but could not yet reproduce structures from experimental measurements
-
1
The deep learning model SMARTERS reconstructs molecular geometries from simulated tip-enhanced Raman spectroscopy images.
-
2
SMARTERS converts hyperspectral TERS data into two-dimensional maps of atomic positions, potentially reducing manual interpretation.
-
3
The model achieved a mean Dice similarity coefficient of 0.842 for atomic-position prediction on the test set.
-
4
Performance declined for non-planar molecules due to weaker signals from atoms farther from the scanning probe tip.
-
5
SMARTERS struggled with experimental TERS images, failing to predict atomic positions due to differences from simulations.
