Retrieving genuine nonlinear Raman responses in ultrafast spectroscopy via deep learning

TL;DR

This paper introduces a deep learning-based frequency-domain denoiser that effectively extracts genuine nonlinear Raman signals from noisy spectroscopic data, improving the analysis of complex vibrational features in ultrafast spectroscopy.

Contribution

It presents a novel convolutional neural network approach trained on simulated data to remove noise and background in nonlinear Raman spectroscopy, outperforming heuristic methods.

Findings

Successfully retrieves asymmetric peaks in stimulated Raman spectra.

Effectively removes noise and background from experimental data.

Applicable to organic molecules and heme proteins.

Abstract

Noise manifests ubiquitously in nonlinear spectroscopy, where multiple sources contribute to experimental signals generating interrelated unwanted components, from random point-wise fluctuations to structured baseline signals. Mitigating strategies are usually heuristic, depending on subjective biases like the setting of parameters in data analysis algorithms and the removal order of the unwanted components. We propose a data-driven frequency-domain denoiser based on a convolutional neural network with kernels of different sizes acting in parallel to extract authentic vibrational features from nonlinear background in noisy spectroscopic raw data. We test our approach by retrieving asymmetric peaks in stimulated Raman spectroscopy (SRS), an ideal test-bed due to its intrinsic complex spectral features combined with a strong background signal. By using a theoretical perturbative toolbox, we efficiently train the network with simulated datasets resembling the statistical properties and lineshapes of the experimental spectra. The developed algorithm is successfully applied to experimental data to obtain noise- and background-free SRS spectra of organic molecules and prototypical heme proteins.