Developing Fully Automated Quality Control Methods for Preprocessing Raman Spectra of Biomedical and Biological Samples

Enhancing real-time cell culture monitoring: Automated Raman model optimization with Taguchi method

Raman spectroscopy has found widespread usage in monitoring cell culture processes both in research and practical applications. However, commonly, preprocessing methods, spectral regions, and modeling parameters have been chosen based on experience or trial-and-error strategies. These choices can significantly impact the performance of the models. There is an urgent need for a simple, effective, and automated approach to determine a suitable procedure for constructing accurate models. This paper introduces the adoption of a design of experiment (DoE) method to optimize partial least squares models for measuring the concentration of different components in cell culture bioreactors. The experimental implementation utilized the orthogonal test table L25(56). Within this framework, five factors were identified as control variables for the DoE method: the window width of Savitzky-Golay smoothing, the baseline correction method, the order of preprocessing steps, spectral regions, and the number of latent variables. The evaluation method for the model was considered as a factor subject to noise. The optimal combination of levels was determined through the signal-to-noise ratio response table employing Taguchi analysis. The effectiveness of this approach was validated through two cases, involving different cultivation scales, different Raman spectrometers, and different analytical components. The results consistently demonstrated that the proposed approach closely approximated the global optimum, regardless of data set size, predictive components, or the brand of Raman spectrometer. The performance of models recommended by the DoE strategy consistently surpassed those built using raw data, underscoring the reliability of models generated through this approach. When compared to exhaustive all-combination experiments, the DoE approach significantly reduces calculation times, making it highly practical for the implementation of Raman spectroscopy in bioprocess monitoring.

Rapid Vector-Based Peak Fitting and Resolution Enhancement for Correlation Analyses of Raman Hyperspectra

Spectroscopic peak parameters are important since they provide information about the analyte under study. Besides obtaining these parameters, peak fitting also resolves overlapped peaks. Thus, the obtained parameters should permit the construction of a higher-resolution version of the original spectrum. However, peak fitting is not an easy task due to computational reasons and because the true nature of the analyte is often unknown. These difficulties are major impediments when large hyperspectral data sets need to be processed rapidly, such as for manufacturing process control. We have developed a novel and relatively fast two-part algorithm to perform peak fitting and resolution enhancement on such data sets. In the first part of the algorithm, estimates of the total number of bands and their parameters were obtained from a representative spectrum in the data set, using a combination of techniques. Starting with these parameter estimates, all the spectra were then iteratively and rapidly fitted with Gaussian bands, exploiting intrinsic features of the Gaussian distribution with vector operations. The best fits for each spectrum were retained. By reducing the obtained bandwidths and commensurately increasing their amplitudes, high-resolution spectra were constructed that greatly improved correlation-based analyses. We tested the performance of the algorithm on synthetic spectra to confirm that this method could recover the ground truth correlations between highly overlapped peaks. To assess effective peak resolution, the method was applied to low-resolution spectra of glucose and compared to results from high-resolution spectra. We then processed a larger spectral data set from mammalian cells, fixed with methanol or air drying, to demonstrate the resolution enhancement of the algorithm on complex spectra and the effects of resolution-enhanced spectra on two-dimensional correlation spectroscopy and principal component analyses. The results indicated that the algorithm would allow users to obtain high-resolution spectra relatively fast and permit the recovery of important aspects of the data's intrinsic correlation structure.

Saline dry fixation for improved cell composition analysis using Raman spectroscopy

Raman spectroscopy enables the label-free assessment of cellular composition. While live cell analysis is the most accurate approach for cellular Raman spectroscopy, the analysis of fixed cells has proved to be very useful, particularly in collaborative projects where samples need to be serially examined by different laboratories or stored and reanalyzed at a later date. However, many chemicals that are widely used for cell fixation directly affect cellular biomolecules, yielding Raman spectra with missing or altered information. In this article, we compared the suitability of dry-fixation with saline versus chemical fixatives. We compared the Raman spectroscopy of saline dry-fixed cells with the more commonly used formaldehyde and methanol fixation and found that dry-fixed cell spectra preserved more cellular information than either chemical fixative. We also assessed the stability of dry-fixed cells over time and found that they were stable for at least 5 months. Finally, a comparison of dry-fixed and live cell spectra revealed effects due to the hydration state of the cells since they were recovered upon rehydrating dry-fixed samples. Thus, for fixed cell Raman spectroscopy, we recommend dry-fixation with unbuffered saline as a superior method to formaldehyde or methanol fixation.

Machine Learning-Assisted Sampling of Surfance-Enhanced Raman Scattering (SERS) Substrates Improve Data Collection Efficiency

Surface-enhanced Raman scattering (SERS) is a powerful technique for sensitive label-free analysis of chemical and biological samples. While much recent work has established sophisticated automation routines using machine learning and related artificial intelligence methods, these efforts have largely focused on downstream processing (e.g., classification tasks) of previously collected data. While fully automated analysis pipelines are desirable, current progress is limited by cumbersome and manually intensive sample preparation and data collection steps. Specifically, a typical lab-scale SERS experiment requires the user to evaluate the quality and reliability of the measurement (i.e., the spectra) as the data are being collected. This need for expert user-intuition is a major bottleneck that limits applicability of SERS-based diagnostics for point-of-care clinical applications, where trained spectroscopists are likely unavailable. While application-agnostic numerical approaches (e.g., signal-to-noise thresholding) are useful, there is an urgent need to develop algorithms that leverage expert user intuition and domain knowledge to simplify and accelerate data collection steps. To address this challenge, in this work, we introduce a machine learning-assisted method at the acquisition stage. We tested six common algorithms to measure best performance in the context of spectral quality judgment. For adoption into future automation platforms, we developed an open-source python package tailored for rapid expert user annotation to train machine learning algorithms. We expect that this new approach to use machine learning to assist in data acquisition can serve as a useful building block for point-of-care SERS diagnostic platforms.

Critical Evaluation of Spectral Resolution Enhancement Methods for Raman Hyperspectra

Overlapping peaks in Raman spectra complicate the presentation, interpretation, and analyses of complex samples. This is particularly problematic for methods dependent on sparsity such as multivariate curve resolution and other spectral demixing as well as for two-dimensional correlation spectroscopy (2D-COS), multisource correlation analysis, and principal component analysis. Though software-based resolution enhancement methods can be used to counter such problems, their performances often differ, thereby rendering some more suitable than others for specific tasks. Furthermore, there is a need for automated methods to apply to large numbers of varied hyperspectral data sets containing multiple overlapping peaks, and thus methods ideally suitable for diverse tasks. To investigate these issues, we implemented three novel resolution enhancement methods based on pseudospectra, over-deconvolution, and peak fitting to evaluate them along with three extant methods: node narrowing, blind deconvolution, and the general-purpose peak fitting program Fityk. We first applied the methods to varied synthetic spectra, each consisting of nine overlapping Voigt profile peaks. Improved spectral resolution was evaluated based on several criteria including the separation of overlapping peaks and the preservation of true peak intensities in resolution-enhanced spectra. We then investigated the efficacy of these methods to improve the resolution of measured Raman spectra. High resolution spectra of glucose acquired with a narrow spectrometer slit were compared to ones using a wide slit that degraded the spectral resolution. We also determined the effects of the different resolution enhancement methods on 2D-COS and on chemical contrast image generation from mammalian cell spectra. We conclude with a discussion of the particular benefits, drawbacks, and potential of these methods. Our efforts provided insight into the need for effective resolution enhancement approaches, the feasibility of these methods for automation, the nature of the problems currently limiting their use, and in particular those aspects that need improvement.

Augmented Two-Dimensional Correlation Spectroscopy for the Joint Analysis of Correlated Changes in Spectroscopic and Disparate Sources

Here, we present an augmented form of two-dimensional correlation spectroscopy, that integrates in a single format data from spectroscopic and multiple non-spectroscopic sources for analysis. The integration is affected by augmenting every spectrum in a hyperspectral data set with relevant non-spectroscopic data to permit two-dimensional correlation analysis^(2D-COS) of the ensemble of augmented spectra. A k-means clustering is then applied to the results of the perturbation domain decomposition to determine which Raman peaks cluster with any of the non-spectroscopic data. We introduce and explain the method with the aid of synthetic spectra and synthetic non-spectroscopic data. We then demonstrate this approach with data using Raman spectra from human embryonic stem cell aggregates undergoing directed differentiation toward pancreatic endocrine cells and parallel bioassays of hormone mRNA expression and C-peptide levels in spent medium. These pancreatic endocrine cells generally contain insulin or glucagon. Insulin has disulfide bonds that produce Raman scattering near 513 cm^-1, but no tryptophan. For insulin-positive cells, we found that the application of multisource correlation analysis revealed a high correlation between insulin mRNA and Raman scattering in the disulfide region. In contrast, glucagon has no disulfide bonds but does contain tryptophan. For glucagon-positive cells, we also observed a high correlation between glucagon mRNA and tryptophan Raman scattering (∼757 cm^-1). We conclude with a discussion of methods to enhance spectral resolution and its effects on the performance of multisource correlation analysis.

Applications of Raman spectroscopy in the development of cell therapies: state of the art and future perspectives

This comprehensive review article discusses current and future perspectives of Raman spectroscopy-based analyses of cell therapy processes and products.