Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated structure discovery for Tip Enhanced Raman Spectroscopy

Harshit Sethi, Markus Junttila, Orlando J Silveira, Adam S Foster

TL;DR

This work presents an encoder-decoder model trained and evaluated on simulated TERS images of planar molecules, enabling direct prediction of molecular structures from spectral simulated data with high accuracy, and demonstrates the feasibility of automating molecular structure identification from TERS images, bypassing traditional manual analysis.

Abstract

Tip-Enhanced Raman Spectroscopy (TERS) provides nanoscale chemical fingerprints alongside high-resolution topographic mapping of molecules, offering a powerful tool for materials discovery. However, TERS image datasets are challenging to interpret and typically demand time-consuming, computationally intensive quantum-chemistry calculations. To overcome this problem, we present an encoder-decoder model trained and evaluated on simulated TERS images of planar molecules, enabling direct prediction of molecular structures from spectral simulated data with high accuracy. Our approach demonstrates the feasibility of automating molecular structure identification from TERS images, bypassing traditional manual analysis. These findings provide a foundation for extending machine learning methods to experimental TERS datasets, potentially accelerating molecular discovery by integrating nanoscale spectroscopy with automated computational analysis.

Automated structure discovery for Tip Enhanced Raman Spectroscopy

TL;DR

This work presents an encoder-decoder model trained and evaluated on simulated TERS images of planar molecules, enabling direct prediction of molecular structures from spectral simulated data with high accuracy, and demonstrates the feasibility of automating molecular structure identification from TERS images, bypassing traditional manual analysis.

Abstract

Tip-Enhanced Raman Spectroscopy (TERS) provides nanoscale chemical fingerprints alongside high-resolution topographic mapping of molecules, offering a powerful tool for materials discovery. However, TERS image datasets are challenging to interpret and typically demand time-consuming, computationally intensive quantum-chemistry calculations. To overcome this problem, we present an encoder-decoder model trained and evaluated on simulated TERS images of planar molecules, enabling direct prediction of molecular structures from spectral simulated data with high accuracy. Our approach demonstrates the feasibility of automating molecular structure identification from TERS images, bypassing traditional manual analysis. These findings provide a foundation for extending machine learning methods to experimental TERS datasets, potentially accelerating molecular discovery by integrating nanoscale spectroscopy with automated computational analysis.
Paper Structure (16 sections, 5 equations, 10 figures, 1 table)

This paper contains 16 sections, 5 equations, 10 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Molecule Dataset
  4. Simulation Method
  5. Image Dataset
  6. Model training
  7. Results
  8. Performance evaluation of SMARTERS
  9. Ablation study on the effect of planarity on the accuracy
  10. Conclusions
  11. Acknowledgements
  12. Model
  13. Dataset Preprocessing
  14. Principal Component Analysis for Planarity Selection and Tip Alignment
  15. Evaluation metrics
  16. ...and 1 more sections

Figures (10)

  • Figure 1: Schematic illustration of the tip-enhanced Raman spectroscopy (TERS) setup, where laser illumination induces localized field enhancement at the SPM tip apex, enabling enhanced Raman signal intensity.
  • Figure 2: Schematic of the deep learning workflow for structure prediction from TERS images.
  • Figure 3: Characteristics of the final dataset of 1,840 preprocessed planar molecules used for simulated TERS image generation.
  • Figure 4: TERS Simulation Pipeline. Our data generation process, adapted from Zhang et al. Zhang2021. (a) The workflow takes a molecule's geometry, calculates DFT properties, simulates interaction, and computes the Raman spectrum. (b) Sum of all TERS Hyperspectral images for visualization.
  • Figure 5: Model performance as a function of planarity. Molecules that deviate more from being flat ($\text{RMSD} > 0.01$ Å) show a noticeable drop in DSC, indicating that the model has more difficulty with non-planar structures.
  • ...and 5 more figures