Machine Learning-Assisted Raman Spectroscopy for pH and Lactate Sensing in Body Fluids

A sensor platform based on SERS detection/janus textile for sweat glucose and lactate analysis toward portable monitoring of wellness status

Herein we report a wearable sweat sensor of a Janus fabric based on surface enhanced Raman scattering (SERS) technology, mainly detecting the two important metabolites glucose and lactate. Janus fabric is composed of electrospinning PU on a piece of medical gauze (cotton), working as the unidirectional moisture transport component (R = 1305%) to collect and transfer sweat efficiently. SERS tags with different structures act as the probe to recognize and detect the glucose and lactate in high sensitivity. Core-shell structured gold nanorods with DTNB inside (AuNRs@DTNB@Au) are used to detect lactate, while gold nanorods with MPBA (AuNRs@MPBA) are used to detect glucose. Through the characteristic SERS information, two calibration functions were established for the concentration determination of glucose and lactate. The concentrations of glucose and lactate in sweat of a 23 years volunteer during three-stage interval running are tested to be 95.5, 53.2, 30.5 μM and 4.9, 13.9, 10.8 mM, indicating the glucose (energy) consumption during exercise and the rapid accumulation of lactate at the early stage accompanied by the subsequent relief. As expected, this sensing system is able to provide a novel strategy for effective acquisition and rapid detection of essential biomarkers in sweat.

Power of Light: Raman Spectroscopy and Machine Learning for the Detection of Lung Cancer

Lung cancer is the leading cause of cancer-related deaths worldwide, emphasizing the urgent need for reliable and efficient diagnostic methods. Conventional approaches often involve invasive procedures and can be time-consuming and costly, thereby delaying the effective treatment. The current study explores the potential of Raman spectroscopy, as a promising noninvasive technique, by analyzing human blood plasma samples from lung cancer patients and healthy controls. In a benchmark study, 16 machine learning models were evaluated by employing four strategies: the combination of dimensionality reduction with classifiers; application of feature selection prior to classification; stand-alone classifiers; and a unified predictive model. The models showed different performances due to the inherent complexity of the data, achieving accuracies from 0.77 to 0.85 and areas under the curve for receiver operating characteristics from 0.85 to 0.94. Hybrid methods incorporating dimensionality reduction and feature selection algorithms present the highest figures of merit. Nevertheless, all machine learning models deliver creditable scores and demonstrate that Raman spectroscopy represents a powerful method for future in vitro diagnostics of lung cancer.

The Good pH probe: non-invasive pH in-line monitoring using Good buffers and Raman spectroscopy

In bioprocesses, the pH value is a critical process parameter that requires monitoring and control. For pH monitoring, potentiometric methods such as pH electrodes are state of the art. However, they are invasive and show measurement value drift. Spectroscopic pH monitoring is a non-invasive alternative to potentiometric methods avoiding this measurement value drift. In this study, we developed the Good pH probe, which is an approach for spectroscopic pH monitoring in bioprocesses with an effective working range between pH 6 and pH 8 that does not require the estimation of activity coefficients. The Good pH probe combines for the first time the Good buffer 3-(N-morpholino)propanesulfonic acid (MOPS) as pH indicator with Raman spectroscopy as spectroscopic technique, and Indirect Hard Modeling (IHM) for the spectral evaluation. During a detailed characterization, we proved that the Good pH probe is reversible, exhibits no temperature dependence between 15 and 40 °C, has low sensitivity to the ionic strength up to 1100 mM, and is applicable in more complex systems, in which other components significantly superimpose the spectral features of MOPS. Finally, the Good pH probe was successfully used for non-invasive pH in-line monitoring during an industrially relevant enzyme-catalyzed reaction with a root mean square error of prediction (RMSEP) of 0.04 pH levels. Thus, the Good pH probe extends the list of critical process parameters monitorable using Raman spectroscopy and IHM by the pH value.

Characterizing viral samples using machine learning for Raman and absorption spectroscopy

Machine learning methods can be used as robust techniques to provide invaluable information for analyzing biological samples in pharmaceutical industries, such as predicting the concentration of viral particles of interest in biological samples. Here, we utilized both convolutional neural networks (CNNs) and random forests (RFs) to predict the concentration of the samples containing measles, mumps, rubella, and varicella-zoster viruses (ProQuad®) based on Raman and absorption spectroscopy. We prepared Raman and absorption spectra data sets with known concentration values, then used the Raman and absorption signals individually and together to train RFs and CNNs. We demonstrated that both RFs and CNNs can make predictions with R2 values as high as 95%. We proposed two different networks to jointly use the Raman and absorption spectra, where our results demonstrated that concatenating the Raman and absorption data increases the prediction accuracy compared to using either Raman or absorption spectrum alone. Additionally, we further verified the advantage of using joint Raman-absorption with principal component analysis. Furthermore, our method can be extended to characterize properties other than concentration, such as the type of viral particles.

Photonic Technology for In Vivo Monitoring of Hypoxia-Ischemia

Surveillance of physiological parameters of newborns during delivery triggers medical decision-making, can rescue life and health, and helps avoid unnecessary cesareans. Here, the development of a photonic technology for monitoring perinatal asphyxia is presented and validated in vivo in a preclinical stage. Contrary to state of the art, the technology provides continuous data in real-time in a non-invasive manner. Moreover, the technology does not rely on a single parameter as pH or lactate, instead monitors changes of the entirety of physiological parameters accessible by Raman spectroscopy. By a fiber-coupled Raman probe that is in controlled contact with the skin of the subject, near-infrared Raman spectra are measured and analyzed by machine learning algorithms to develop classification models. As a performance benchmarking, various hybrid and non-hybrid classifiers are tested. In an asphyxia model in newborn pigs, more than 1000 Raman spectra are acquired at three different clinical phases-basal condition, hypoxia-ischemia, and post-hypoxia-ischemia stage. In this preclinical proof-of-concept study, figures of merit reach 90% levels for classifying the clinical phases and demonstrate the power of the technology as an innovative medical tool for diagnosing a perinatal adverse outcome.

Fast Synthesis of Au Nanoparticles on Metal-Phenolic Network for Sweat SERS Analysis

The biochemical composition of sweat is closely related to the human physiological state, which provides a favorable window for the monitoring of human health status, especially for the athlete. Herein, an ultra-simple strategy based on the surface-enhanced Raman scattering (SERS) technique for sweat analysis is established. Metal-phenolic network (MPN), an outstanding organic-inorganic hybrid material, is adopted as the reductant and platform for the in situ formation of Au-MPN, which displays excellent SERS activity with the limit of detection to 10^-15 M for 4-mercaptobenzoic acid (4-MBA). As an ultrasensitive SERS sensor, Au-MPN is capable of discriminating the molecular fingerprints of sweat components acquired from a volunteer after exercise, such as urea, uric acid, lactic acid, and amino acid. For pH sensing, Au-MPN/4-MBA efficiently presents the pH values of the volunteer's sweat, which can indicate the electrolyte metabolism during exercise. This MPN-based SERS sensing strategy unlocks a new route for the real-time physiological monitoring of human health.

Design and Development of a Bimodal Optical Instrument for Simultaneous Vibrational Spectroscopy Measurements

Vibrational spectroscopy techniques are widely used in analytical chemistry, physics and biology. The most prominent techniques are Raman and Fourier-transform infrared spectroscopy (FTIR). Combining both techniques delivers complementary information of the test sample. We present the design, construction, and calibration of a novel bimodal spectroscopy system featuring both Raman and infrared measurements simultaneously on the same sample without mutual interference. The optomechanical design provides a modular flexible system for solid and liquid samples and different configurations for Raman. As a novel feature, the Raman module can be operated off-axis for optical sectioning. The calibrated system demonstrates high sensitivity, precision, and resolution for simultaneous operation of both techniques and shows excellent calibration curves with coefficients of determination greater than 0.96. We demonstrate the ability to simultaneously measure Raman and infrared spectra of complex biological material using bovine serum albumin. The performance competes with commercial systems; moreover, it presents the additional advantage of simultaneously operating Raman and infrared techniques. To the best of our knowledge, it is the first demonstration of a combined Raman-infrared system that can analyze the same sample volume and obtain optically sectioned Raman signals. Additionally, quantitative comparison of confocality of backscattering micro-Raman and off-axis Raman was performed for the first time.

Identification of Hypoxia-Ischemia by chemometrics considering systemic changes of the physiology

Perinatal asphyxia represents a major medical disorder and is related to around a fourth of all neonatal deaths worldwide. Specific thresholds for lactate or pH levels define the gold standard for detecting hypoxic-ischemic events as physiological abnormalities. In contrast to current gold standard, we analyze the systemic picture, represented by the whole set of biochemical parameters from blood gas analysis, by multiparametric machine learning algorithms. In a swine model with 22 objects, we investigate the impact of neonatal hypoxic-ischemic encephalopathy on 18 individual physiological parameters. In a first approach, we study the statistical significance of individual parameters by univariate analysis methods. In a second approach, we take the most relevant parameters as input for the development of predictive models by different hybrid and non hybrid classification algorithms. The predictive power of our multiparametric models outperforms by far the limited performance of pH and lactate as reliable indicators, despite strong correlation with hypoxic-ischemic events. We have been able to detect hypoxic-ischemic events even one hour after the episode, with accuracies close to 100% in contrast to pH or lactate-based diagnosis with 62% and 78%, respectively. By all machine learning algorithms, lactate is recognized as the main contributor due to its longer-term evidence of hypoxia-ischemia episodes. However, substantial improvement of the diagnosis is achieved by predictions based on a systemic picture of different physiological parameters. Our results prove the potential applicability of our method as a support tool for decision-making that will allow obstetricians to identify hypoxic ischemic episodes more accurately during labor.

Identifying the charge density and dielectric environment of graphene using Raman spectroscopy and deep learning

We built a CNN model to classify graphene Raman spectra. Compared to other deep learning models and machine learning algorithms studied in this work, the CNN model achieves a high accuracy of 99% and is less sensitive to the SNR of Raman spectra.

Fluorescence and Photoacoustic Spectroscopy-Based Assessment of Mitochondrial Dysfunction in Oral Cancer Together with Machine Learning: A Pilot Study

The current study reports an integrated approach of machine learning and tryptophan fluorescence and photoacoustic spectral properties to assess the mitochondrial status under oral pathological conditions. The mitochondria in the study were isolated from oral cancer tissues and adjacent normal counterparts, and the corresponding fluorescence and photoacoustic spectra of tryptophan were recorded at 281 nm pulsed laser excitations. A set of features were selected from the pre-processed spectra and were used to classify the data using support vector machine (SVM) learning in the MATLAB platform. SVM analysis demonstrated clear differentiation between mitochondria isolated from normal and cancer tissues for fluorescence (sensitivity, 86.6%; specificity, 90%) and photoacoustic (sensitivity, 86.6%; specificity, 96.6%) measurements. Further investigation into the influence of change in protein conformation on the nature of tryptophan spectral properties was evaluated by 8-anilino-1-naphthalene sulfonic acid (ANS) fluorescence assay. The impact of protein structural changes on the mitochondrial functions was also estimated by mitochondrial membrane potential (MMP), reactive oxygen species (ROS), and cytochrome c oxidase (COX) assays, suggesting an altered mitochondrial function. The findings indicate that tryptophan fluorescence and photoacoustic spectral properties together with machine learning algorithms may delineate the mitochondrial functional status in vitro, indicating its translational potential.

Machine learning algorithms to control concentrations of carbon nanocomplexes in a biological medium via optical absorption spectroscopy: how to choose and what to expect?

A solution of spectroscopic inverse problems, implying determination of target parameters of the research object via analysis of spectra of various origins, is an overly complex task, especially in case of strong variability of the research object. One of the most efficient approaches to solve such tasks is use of machine learning (ML) methods, which consider some unobvious information relevant to the problem that is present in the data. Here, we compare ML approaches to the problem of nanocomplex concentrations determination in human urine via optical absorption spectra, perform preliminary analysis of the data array, find optimal parameters for several of the most popular ML methods, and analyze the results.