Bayesian modelling and quantification of Raman spectroscopy

Matthew T. Moores, Kirsten Gracie, Jake Carson, Karen Faulds, Duncan Graham, Mark Girolami2016 · DOI

Abstract

Raman spectroscopy can be used to identify molecules such as DNA by the characteristic scattering of light from a laser. It is sensitive at very low concentrations and can accurately quantify the amount of a given molecule in a sample. The presence of a large, nonuniform background presents a major challenge to analysis of these spectra. To overcome this challenge, we introduce a sequential Monte Carlo (SMC) algorithm to separate the observed spectrum into a series of peaks plus a smoothly-varying baseline, corrupted by additive white noise. The peaks are modelled using Lorentzian or Gaussian broadening functions, while the baseline is estimated using a penalised cubic spline. This latent continuous representation accounts for differences in resolution between measurements. By incorporating this representation in a Bayesian model, we can quantify the relationship between molecular concentration and peak intensity, thereby providing an improved estimate of the limit of detection (LOD), which is of major importance in analytical chemistry.

Introduction: The Challenge of Spectral Backgrounds

Raman spectroscopy is a cornerstone of molecular identification, utilizing the unique "fingerprint" of spectral peaks created when laser light scatters off molecules. However, raw Raman data is frequently obscured by a large, non-uniform background—often caused by fluorescence—that can significantly exceed the intensity of the signal peaks themselves.

A raw Raman spectrum showing significant background intensity and characteristic peaks.

Traditionally, researchers address this through "baseline subtraction" as a pre-processing step. However, this approach typically treats the baseline as a known, fixed quantity, failing to account for the uncertainty in its estimation. At low molecular concentrations, where the signal-to-noise ratio is poor, this neglect can lead to biased results and inaccurate quantification.

A Unified Bayesian Framework

To address these limitations, we propose a robust Bayesian framework that jointly estimates signal peaks and the background baseline. The model decomposes the Raman spectrum into three distinct components:

  1. The Baseline: Represented as a smoothly varying continuous function using penalized B-splines. This allows the model to capture background fluorescence without interfering with the sharper signal peaks.
  2. The Signal: Modeled as an additive mixture of physics-based broadening functions—either Gaussian (representing Doppler broadening) or Lorentzian (representing collisional broadening).
  3. The Noise: Modeled as additive white noise with constant variance.

By treating the baseline and peaks simultaneously, the model avoids the artifacts introduced by heuristic pre-processing.

Sequential Monte Carlo (SMC) for Uncertainty Quantification

To navigate the complex parameter space of this model, we employ a Sequential Monte Carlo (SMC) algorithm utilizing likelihood tempering. Unlike standard optimization techniques that provide a single "best fit," SMC evolves a population of "particles" to map the entire posterior distribution. This provides a rigorous quantification of uncertainty for all parameters, including peak intensity, width, and the baseline itself. The implementation is available as an open-source R package, serrsBayes.

Results: Enhancing Sensitivity and Detection Limits

The model was evaluated using a dilution study of TAMRA (tetramethylrhodamine) dye via Surface-Enhanced Raman Scattering (SERS), comprising 315 spectra across concentrations from 0.13 nM to 24.7 nM.

Baseline-corrected spectra and posterior mean fits for TAMRA at low concentrations (0.13 nM and 0.65 nM).

A key advantage of this framework is its ability to perform Multivariate Calibration (MVC). By analyzing all 32 peaks of the TAMRA molecule simultaneously rather than focusing on a single dominant peak, the model achieves significantly higher sensitivity.

Our Bayesian MVC approach predicted a Limit of Detection (LOD) for TAMRA between 6 and 16 picomolar (pM). In comparison, a traditional univariate analysis of the same dataset estimated the LOD at 99.5 pM. This represents nearly an order of magnitude improvement in sensitivity. Furthermore, the model successfully produced 95% Highest Posterior Density (HPD) intervals, providing clear statistical confidence for the chemical analysis.

Conclusion and Future Directions

The proposed Bayesian framework offers a significant advancement in the quantification of Raman spectroscopy, particularly at the low concentrations critical for analytical chemistry. While the current model assumes a linear relationship between concentration and intensity, future work will explore non-linear interactions, such as preferential attachment in SERS. Additionally, the researchers aim to extend the model to "multiplexed" spectra containing multiple molecular types and investigate more computationally efficient Bayesian methods to handle larger datasets.

Summary updated

Abstract

Raman spectroscopy can be used to identify molecules such as DNA by the characteristic scattering of light from a laser. It is sensitive at very low concentrations and can accurately quantify the amount of a given molecule in a sample. The presence of a large, nonuniform background presents a major challenge to analysis of these spectra. To overcome this challenge, we introduce a sequential Monte Carlo (SMC) algorithm to separate each observed spectrum into a series of peaks plus a smoothly-varying baseline, corrupted by additive white noise. The peaks are modelled as Lorentzian, Gaussian, or pseudo-Voigt functions, while the baseline is estimated using a penalised cubic spline. This latent continuous representation accounts for differences in resolution between measurements. The posterior distribution can be incrementally updated as more data becomes available, resulting in a scalable algorithm that is robust to local maxima. By incorporating this representation in a Bayesian hierarchical regression model, we can quantify the relationship between molecular concentration and peak intensity, thereby providing an improved estimate of the limit of detection, which is of major importance to analytical chemistry.
BibTeX
@inproceedings{moores2016bayesian,
  author = {Matthew T. Moores and Kirsten Gracie and Jake Carson and Karen Faulds and Duncan Graham and Mark Girolami},
  title = {{Bayesian modelling and quantification of Raman spectroscopy}},
  booktitle = {Strathprints:},
  year = {2016},
  doi = {10.48550/arxiv.1604.07299},
}
Figure