Large Spectral Data Sets Research Articles

TinyML-Raman: A novel IoT based field-deployable spectra analysis for accurate identification of pharmaceuticals and trace dye-pesticide mixtures from facile SERS method

BackgroundUpcoming inexpensive, compact Internet of Things (IoT) microcontrollers i.e., tiny-machine learning (TinyML) takes the ML driven Raman spectroscopy one step ahead for realization of more affordable and highly compact field deployable instruments. Further, lack of large spectral datasets and need for numerous specialized SERS substrates impede the development of various ML-based surface enhanced Raman spectroscopy (SERS) applications. The aim is to introduce TinyML analysis on wide range of spectra classes using customized dataset obtained with low-cost SERS. In this regard, it is vital to establish an optimum ML model and efficient data handling methodology for low memory TinyML units. ResultsWe introduce a novel TinyML methodology for accurate classification of large spectra classes with smartphone assistance for data communication and results visualization. To generate large customized spectral dataset, we present a facile, micro-drop SERS using Au colloids and reusable grooved Al substrates. The results demonstrated that memory efficient 8-bit data quantization based convolutional neural network is effective for accurate identification of 22 different spectra classes of trace dye-pesticide mixtures and pharmaceuticals. In this novel quantized data analysis on significantly varied intensity and complex variation spectra classes i.e., many individual, binary-mixtures and some with varied compositions, data normalization is shown to be powerful for improving ML classification accuracy from about 55 % to >99.5 %. Its robustness is demonstrated using inter-instrument driven data variations such as spectral shifts, high noise, and abscissa-flip, with five-fold cross validation of model performance. In addition, on-site quantification of analyte through spectral intensity is also demonstrated. SignificanceThis study opens up a new approach of ML analysis towards realization of next generation field deployable analytical instruments maintaining data privacy. It presents a detailed procedure of quantized spectral data analysis and its implementation in TinyML, attractive for various users and instrument manufacturers. The presented innovative computer-free ML analysis can be employed in all types of spectrometers, meeting the common goal of Raman spectroscopy i.e., accurate identification of complex spectra classes.

Machine Learning-Driven SERS Nanoendoscopy and Optophysiology.

A frontier of analytical sciences is centered on the continuous measurement of molecules in or near cells, tissues, or organs, within the biological context in situ, where the molecular-level information is indicative of health status, therapeutic efficacy, and fundamental biochemical function of the host. Following the completion of the Human Genome Project, current research aims to link genes to functions of an organism and investigate how the environment modulates functional properties of organisms. New analytical methods have been developed to detect chemical changes with high spatial and temporal resolution, including minimally invasive surface-enhanced Raman scattering (SERS) nanofibers using the principles of endoscopy (SERS nanoendoscopy) or optical physiology (SERS optophysiology). Given the large spectral data sets generated from these experiments, SERS nanoendoscopy and optophysiology benefit from advances in data science and machine learning to extract chemical information from complex vibrational spectra measured by SERS. This review highlights new opportunities for intracellular, extracellular, and in vivo chemical measurements arising from the combination of SERS nanosensing and machine learning.

Spectral Image Reconstruction Using Recovered Basis Vector Coefficients

Spectral imaging plays a crucial role in various fields, including remote sensing, medical imaging, and material analysis, but it often requires specialized and expensive equipment, making it inaccessible to many. Its application is also limited by the interdependent constraints of temporal, spatial, and spectral resolutions. In order to address these issues, and thus, obtain high-quality spectral images in a time-efficient and affordable manner, we proposed one two-step method for spectral image reconstruction from easily available RGB images under the down-sampling schemes. Specifically, we investigated how RGB values characterize spectral reflectance and found that, compared to the intuitive and straightforward RGB images themselves, their corresponding basis vector coefficients can represent the prior information of spectral images more explicitly and are better suited for spectral image reconstruction tasks. Thus, we derived one data-driven algebraic method to recover the corresponding basis vector coefficients from RGB images in an analytical form and then employed one CNN-based neural network to learn the patch-level mapping from the recovered basis vector coefficients to spectral images. To evaluate the effect of introducing the basis vector coefficient recovery step, several CNNs which typically perform well in spectral image reconstruction are chosen as benchmarks to compare the variation in reconstruction performance. Experimental results on a large public spectral image dataset and our real-world dataset demonstrate that compared to the unaltered version, those CNNs guided by the recovered basis vector coefficients can achieve significant performance improvement in the reconstruction accuracy. Furthermore, this method is plug-and-play, with very little computational performance consumption, thus maintaining a high speed of calculation.

A balancing act: Synthetic oversampling corrects skewed distribution of aflatoxin in maize and improves prediction of contaminated kernels

Current spectral models for detecting aflatoxin (AF) in maize kernels are limited by very low representation of kernels with AF levels above the models’ specified contamination thresholds. With the aim of improving prediction model sensitivity to AF-contaminated kernels, two methods were evaluated for artificially enriching the representation of kernels containing ≥150 parts per billion (ppb) AF in a large spectral dataset: humid incubation and synthetic oversampling. The addition of synthetic training samples improved prediction accuracy of kernels containing ≥150 ppb AF from 51% to 80%. Humid incubation contributed additional kernels with intermediate levels of AF contamination (primarily 5–75 ppb), but did not change the overall distribution of AF contamination. Feature importance distributions overlapped among models at 329–345 nm, 380–385.5 nm, 415–425 nm, 639–668 nm, and 1013.5–1060 nm. These spectral ranges could be applicable to the development of limited wavelength grain sorting devices for low-resource settings.

Full spectrum flow cytometry in the clinical laboratory.

Contemporary full spectrum or "spectral" flow cytometry is a recently developed technology that allows for high-dimensional flow cytometric analyses of cells and particles in suspension. This single-cell technology has gained popularity in research settings because it can conservatively detect 35 or more antigens simultaneously in a single-tube assay format. Recently, spectral flow cytometry has obtained regulatory approval for use as an in vitro diagnostic device in China and Europe, enabling use of this technology in some clinical flow cytometry laboratories. The purpose of this review is to describe the basic principles of conventional and spectral flow cytometry, contrasting these two technologies. To illustrate the analytic power of spectral flow cytometry, we provide an example of spectral flow cytometry data analyses and the use of a machine learning algorithm to harvest the vast amount of information contained within large spectral flow cytometry datasets. Finally, we discuss the advantages of spectral flow cytometry adoption in clinical laboratories and preliminary studies comparing the performance of this technology relative to conventional flow cytometers that are currently used in clinical laboratory environments.

Fast-Training Deep Learning Algorithm for Multiplex Quantification of Mammalian Bioproduction Metabolites via Contactless Short-Wave Infrared Hyperspectral Sensing.

Within the biopharmaceutical sector, there exists the need for a contactless multiplex sensor, which can accurately detect metabolite levels in real time for precise feedback control of a bioreactor environment. Reported spectral sensors in the literature only work when fully submerged in the bioreactor and are subject to probe fouling due to a cell debris buildup. The use of a short-wave infrared (SWIR) hyperspectral (HS) cam era allows for efficient, fully contactless collection of large spectral datasets for metabolite quantification. Here, we report the development of an interpretable deep learning system, a convolution metabolite regression (CMR) approach that detects glucose and lactate concentrations using label-free contactless HS images of cell-free spent media samples from Chinese hamster ovary (CHO) cell growth flasks. Using a dataset of <500 HS images, these CMR algorithms achieved a competitive test root-mean-square error (RMSE) performance of glucose quantification within 27 mg/dL and lactate quantification within 20 mg/dL. Conventional Raman spectroscopy probes report a validation performance of 26 and 18 mg/dL for glucose and lactate, respectively. The CMR system trains within 10 epochs and uses a convolution encoder with a sparse bottleneck regression layer to pick the best-performing filters learned by CMR. Each of these filters is combined with existing interpretable models to produce a metabolite sensing system that automatically removes spurious predictions. Collectively, this work will advance the safe and efficient adoption of contactless deep learning sensing systems for fine control of a variety of bioreactor environments.

Neural network for determining an asteroid mineral composition from reflectance spectra

Context.Chemical and mineral compositions of asteroids reflect the formation and history of our Solar System. This knowledge is also important for planetary defence and in-space resource utilisation. In the next years, space missions will generate extensive spectral datasets from asteroids or planets with spectra that will need to be processed in real time.Aims.We aim to develop a fast and robust neural-network-based method for deriving the mineral modal and chemical compositions of silicate materials from their visible and near-infrared spectra. The method should be able to process raw spectra without significant pre-processing.Methods.We designed a convolutional neural network with two hidden layers for the analysis of the spectra, and trained it using labelled reflectance spectra. For the training, we used a dataset that consisted of reflectance spectra of real silicate samples stored in the RELAB and C-Tape databases, namely olivine, orthopyroxene, clinopyroxene, their mixtures, and olivine-pyroxene-rich meteorites.Results.We used the model on two datasets. First, we evaluated the model reliability on a test dataset where we compared the model classification with known compositional reference values. The individual classification results are mostly within 10 percentage-point intervals around the correct values. Second, we classified the reflectance spectra of S-complex (Q-type and V-type, also including A-type) asteroids with known Bus–DeMeo taxonomy classes. The predicted mineral chemical composition of S-type and Q-type asteroids agree with the chemical composition of ordinary chondrites. The modal abundances of V-type and A-type asteroids show a dominant contribution of orthopyroxene and olivine, respectively. Additionally, our predictions of the mineral modal composition of S-type and Q-type asteroids show an apparent depletion of olivine related to the attenuation of its diagnostic absorptions with space weathering. This trend is consistent with previous results of the slower pyroxene response to space weathering relative to olivine.Conclusions.The neural network trained with real silicate samples and their mixtures is applicable for a quantitative mineral evaluation of spectra of asteroids that are rich in dry silicates. The modal abundances and mineral chemistry of common silicates (olivine and pyroxene) can be derived with an accuracy better than 10 percentage points. The classification is fast and has a relatively small computer-memory footprint. Therefore, our code is suitable for processing large spectral datasets in real time.

Relativistic Ritz approach to hydrogenlike atoms: Theoretical considerations

The Rydberg formula along with the Ritz quantum defect ansatz has been a standard theoretical tool used in atomic physics since before the advent of quantum mechanics, yet this approach has remained limited by its non-relativistic foundation. Here I present a long-distance relativistic effective theory describing hydrogen-like systems with arbitrary mass ratios, thereby extending the canonical Ritz-like approach. Fitting the relativistic theory to the hydrogen energy levels predicted by bound-state QED indicates that it is superior to the canonical, nonrelativistic approach. An analytic analysis reveals nonlinear consistency relations within the bound-state QED level predictions that relate higher-order corrections to those at lower order, providing guideposts for future perturbative calculations as well as insights into the asymptotic behavior of Bethe logarithms. Applications of the approach include fitting to atomic spectroscopic data, allowing for the determination the fine-structure constant from large spectral data sets and also to check for internal consistency of the data independently from bound-state QED.

Gamma-ray anomaly detector for airborne surveys based on a machine learning methodology

Machine learning methods were used to develop an automated classification model for detecting gamma-ray spectra with anomalous (i.e., non-background) signatures collected during airborne surveys. Spectra were preprocessed with an altitude-based normalization procedure, followed by application of a digital filter to remove baseline and noise effects. Segments of the preprocessed spectra were then used as input patterns to piecewise linear discriminant analysis (PLDA). For use in building the classification model, a training set was constructed based on two data classes: (1) spectra containing various radioisotope signatures and (2) background spectra. Through the use of a piecewise linear discriminant based on seven separating boundaries, a general-purpose spectral anomaly detector was developed for use in the automated screening of large spectral datasets collected during airborne surveys. The intended application of the methodology is to aid first responders in locating lost or stolen radioactive sources or in managing other incidents in which radioactive material is released into the environment. In developing and testing the methodology, laboratory spectra, spectra collected during airborne surveys, and mathematically synthesized spectra were used. When applied to 17 airborne surveys in which known radioisotope sources were present, the anomaly detector was able to locate each source with high confidence. While false detections were observed at a rate of 3.7%, many of these were in the vicinity of the known source location but the radioisotope signature could not be visually observed in the spectrum. For false detections away from the source locations, elevated signatures from naturally occurring background components were typically observed.

A hybrid model to predict nitrogen concentration in heterogeneous grassland using field spectroscopy

Field spectroscopy is a rapid and non-destructive tool used for the estimation of nitrogen concentration (N%) of vegetation. Empirical and physically-based models are widely used for retrieving N%. However, model transferability to different times and locations, and feature redundancy remain the two key challenges of field spectroscopy analysis. Here we addressed these problems by developing a hybrid method (i.e., a combination of physically-based (PROSAIL) and empirical models) to retrieve N% in grasslands. We used a large spectral dataset with >6000 samples collected over 8 years (2009–2016) for grassland farms across New Zealand. The hybrid model combines the features derived from PROSAIL inversion and an empirical model and develops a predictive model using a Gaussian Process Regression (GPR) algorithm. The model performance is tested on spatially and temporally independent data and compared with PROSAIL and empirical models. The hybrid model achieves a higher performance with an RMSE (%N), R2 and Mean Prediction Interval Width (MPIW) of 0.27, 0.78 and 0.26 as compared to empirical (RMSE = 0.28, R2 = 0.77 and MPIW = 0.32) and physically-based models (RMSE = 0.33, R2 = 0.65 and MPIW = 0.56). In addition, the hybrid model significantly outperforms the physically-based and empirical models during autumn (RMSE = 0.32, R2 = 0.78 and MPIW = 0.11) and summer (RMSE = 0.27, R2 = 0.80 and MPIW = 0.16) seasons.

Deep reinforced neural network model for cyto-spectroscopic analysis of epigenetic markers for automated oral cancer risk prediction

Understanding epigenetic changes can provide vital information for early stage oral cancer diagnosis. Vibrational spectroscopy methods like Raman spectroscopy (RS) and Fourier Transform Infrared Spectroscopy (FTIR) can provide several advantages over conventional molecular biology methods, by incorporating information from fingerprint regions. Moreover, application of advanced spectral analysis tools like deep learning (DL) techniques can be efficiently applied for analyzing the large spectral dataset and extracting the vital features. Epigenetic changes are identifies in oral epithelial cells of healthy individuals, oral leucoplakia and squamous cell carcinoma patients through analysis of Raman (400-1800 ​cm−1) and FTIR data (700-2000 ​cm−1). Deep reinforced neural network (DRNN) model is employed to classify the epigenetic changes identified from the Raman and FTIR spectra. Feature extraction layer of DL model uses peak detection layer and reinforced learning layer to identify significant epigenetic features. Classification layer is made up of N numbers of back propagated Artificial Neural Network (ANN) layers. DL model developed is fully automated and overcome the wave shift problem of spectroscopic data. Testing accuracy of the proposed DRNN model is 83.33%. Class wise accuracies for NRML, OLPK and OSCC are 83.3%, 87% and 95.24%, respectively. Proposed DRNN model attains an overall ROC of 0.88 Present study employs combination of two complementary vibrational spectroscopy methods for oral cancer detection and chemometric analysis of the spectral features with DRNN mode. Identification of the epigenetic changes and utilization of the knowledge in cancer prediction will enable the proposed study to develop smart point-of-care diagnostic system.

Lectin-Modified Bacterial Cellulose Nanocrystals Decorated with Au Nanoparticles for Selective Detection of Bacteria Using Surface-Enhanced Raman Scattering Coupled with Machine Learning

Bacterial cellulose nanocrystals (BCNCs) are tunable and biocompatible cellulose nanomaterials that can be easily bioconjugated and used for biosensing applications. We report the application of concanavalin A (con A) lectin-modified BCNCs (con A + BCNCs) for bacterial isolation and label-free surface-enhanced Raman spectroscopy (SERS) detection of bacterial species using Au nanoparticles (AuNPs). The aggregated AuNP + bacteria + (con A + BCNC) conjugates generated SERS hot spots that enabled the SERS detection of the strain Escherichia coli 8739 at the 103 CFU/mL level. The optimized detection assay was then used to differentiate 19 common bacterial strains. The large SERS spectral dataset for the 19 bacterial strains was analyzed using the support vector machine (SVM), an optimization-based machine-learning technique that worked as a binary classifier. The SVM classifier showed a high overall accuracy of 87.7% in correctly discriminating bacterial strains. This study illustrates the potential of combining low-cost nanocellulose-based SERS biosensors with machine-learning techniques for the analysis of large spectral datasets.

Raman spectroscopy is a powerful tool in molecular paleobiology: An analytical response to Alleon etal. (https://doi.org/10.1002/bies.202000295).

A recent article argued that signals from conventional Raman spectroscopy of organic materials are overwhelmed by edge filter and fluorescence artefacts. The article targeted a subset of Raman spectroscopic investigations of fossil and modern organisms and has implications for the utility of conventional Raman spectroscopy in comparative tissue analytics. The inferences were based on circular reasoning centered around the unconventional analysis of spectra from just two samples, one modern, and one fossil. We validated the disputed signals with in situ Fourier-Transform Infrared (FT-IR) Spectroscopy and through replication with different lasers, filters, and operators in independent laboratories. Our Raman system employs a holographic notch filter which is not affected by edge filter or other artefacts. Multiple lines of evidence confirm that conventional Raman spectra of fossils contain biologically and geologically meaningful information. Statistical analyses of large Raman and FT-IR spectral data sets reveal patterns in fossil composition and yield valuable insights into the history of life.

MS2DeepScore: a novel deep learning similarity measure to compare tandem mass spectra

Mass spectrometry data is one of the key sources of information in many workflows in medicine and across the life sciences. Mass fragmentation spectra are generally considered to be characteristic signatures of the chemical compound they originate from, yet the chemical structure itself usually cannot be easily deduced from the spectrum. Often, spectral similarity measures are used as a proxy for structural similarity but this approach is strongly limited by a generally poor correlation between both metrics. Here, we propose MS2DeepScore: a novel Siamese neural network to predict the structural similarity between two chemical structures solely based on their MS/MS fragmentation spectra. Using a cleaned dataset of > 100,000 mass spectra of about 15,000 unique known compounds, we trained MS2DeepScore to predict structural similarity scores for spectrum pairs with high accuracy. In addition, sampling different model varieties through Monte-Carlo Dropout is used to further improve the predictions and assess the model’s prediction uncertainty. On 3600 spectra of 500 unseen compounds, MS2DeepScore is able to identify highly-reliable structural matches and to predict Tanimoto scores for pairs of molecules based on their fragment spectra with a root mean squared error of about 0.15. Furthermore, the prediction uncertainty estimate can be used to select a subset of predictions with a root mean squared error of about 0.1. Furthermore, we demonstrate that MS2DeepScore outperforms classical spectral similarity measures in retrieving chemically related compound pairs from large mass spectral datasets, thereby illustrating its potential for spectral library matching. Finally, MS2DeepScore can also be used to create chemically meaningful mass spectral embeddings that could be used to cluster large numbers of spectra. Added to the recently introduced unsupervised Spec2Vec metric, we believe that machine learning-supported mass spectral similarity measures have great potential for a range of metabolomics data processing pipelines.

Biofluid analysis and classification using IR and 2D-IR spectroscopy

Vibrational spectroscopy has produced valuable information for biomedical research owing to its label-free and high-throughput capabilities. However, the complexity of and large number of variables of spectral datasets has seen the increasing application of multivariate analysis (MVA) and machine learning algorithms in recent years. In particular, the use of these techniques applied to the analysis of IR spectra of biological samples has been demonstrated as a powerful tool for the rapid sample analysis and diagnosis of disease. In this article, we review a variety of classification techniques employed for the analysis of infrared (IR) spectral datasets of biofluids, quoting prediction accuracies to demonstrate their effectiveness. With the advent of new technologies, two-dimensional infrared spectroscopy (2D-IR) has recently been applied to biomedical problems and shows potential future applications in biofluid analysis, however with complex multi-dimensional datasets there is a desire for advanced analytical techniques. As the application of 2D-IR to biofluids and physiological protein samples is in its infancy, large spectral datasets of biofluids suitable for classification are not readily available. It is imperative to establish in what way 2D-IR datasets respond to pre-processing and analytical methods. For the first time we draw on the classification techniques applied to IR datasets discussed in this review and relevant 2D-IR studies to discuss the future of machine learning algorithms in 2D-IR spectroscopy.

Soil properties: Their prediction and feature extraction from the LUCAS spectral library using deep convolutional neural networks

Soil, as a non-renewable resource, should be monitored continuously to prevent its degradation and promote sustainable agriculture. Soil spectroscopy in the visible-near infrared range is a fast and cost-effective analytical technique to predict soil properties. Although traditional machine learning methods are widely used for modeling soil spectral data, large spectral datasets may require better analytical methods for big data. Here, we explored the modeling potential of deep convolutional neural networks (DCNNs) for soil properties based on a large soil spectral library. The European topsoil dataset provided by the Land Use/Cover Area frame Survey (LUCAS) was used for DCNN modeling with the original absorbance spectra. Two single-task 16-layer DCNN models (LucasResNet-16 and LucasVGGNet-16) were used to make regression predictions of seven soil properties and classification predictions of soil texture. The effects of data pre-processing on single-task and multi-task DCNN modeling were assessed. The SHapley Additive exPlanations method was used to interpret the output of a DCNN model (LucasResNet-16). The DCNN models produced accurate predictions for most soil properties, and were superior to a single-task shallow convolutional neural network and traditional machine learning methods. Spectral transformation was effective for predicting some soil properties, while spectral downsampling led to a reduction in the modeling accuracy. The performance of a multi-task DCNN model built on the basis of LucasResNet-16 was improved compared with the performance of the single-task model. Soil organic carbon content, nitrogen content, cation exchange capacity, pH, and calcium carbonate content were well predicted, with the root mean squared error of 19.130 g∙kg−1, 0.971 g∙kg−1, 6.614 cmol(+)∙kg−1, 0.326, and 24.526 g∙kg−1, respectively. The overall classification accuracy of soil texture was 0.749 (four groups) and 0.566 (12 levels). The position of feature wavelengths differed among the soil properties, for which multiple characteristic peaks were common. This study fully demonstrates the modeling potential of deep learning with soil ultraspectral data, which could enhance precision agriculture.

Wheat Kernel Variety Identification Based on a Large Near-Infrared Spectral Dataset and a Novel Deep Learning-Based Feature Selection Method.

Near-infrared (NIR) hyperspectroscopy becomes an emerging nondestructive sensing technology for inspection of crop seeds. A large spectral dataset of more than 140,000 wheat kernels in 30 varieties was prepared for classification. Feature selection is a critical segment in large spectral data analysis. A novel convolutional neural network-based feature selector (CNN-FS) was proposed to screen out deeply target-related spectral channels. A convolutional neural network with attention (CNN-ATT) framework was designed for one-dimension data classification. Popular machine learning models including support vector machine (SVM) and partial least square discrimination analysis were used as the benchmark classifiers. Features selected by conventional feature selection algorithms were considered for comparison. Results showed that the designed CNN-ATT produced a higher performance than the compared classifier. The proposed CNN-FS found a subset of features, which made a better representation of raw dataset than conventional selectors did. The CNN-ATT achieved an accuracy of 93.01% using the full spectra and keep its high precision (90.20%) by training on the 60-channel features obtained via the CNN-FS method. The proposed methods have great potential for handling the analyzing tasks on other large spectral datasets. The proposed feature selection structure can be extended to design other new model-based selectors. The combination of NIR hyperspectroscopic technology and the proposed models has great potential for automatic nondestructive classification of single wheat kernels.

Data consistency and classification model transferability across biomedical Raman spectroscopy systems

AbstractSurgical guidance applications using Raman spectroscopy are being developed at a rapid pace in oncology to ensure safe and complete tumor resection during surgery. Clinical translation of these approaches relies on the acquisition of large spectral and histopathological data sets to train classification models. Data calibration must ensure compatibility across Raman systems and predictive model transferability to allow multi‐centric studies to be conducted. This paper addresses issues relating to Raman measurement standardization by first comparing Raman spectral measurements made on an optical phantom and acquired with nine distinct point probe systems and one wide‐field imaging instrument. Data standardization method led to normalized root‐mean‐square deviations between instruments of 2%. A classification model discriminating between white and gray matter was trained with one point probe system. When used to classify independent data sets acquired with the other systems, model predictions led to &gt;95% accuracy, preliminarily demonstrating model transferability across different biomedical Raman spectroscopy instruments.

Use of Large and Diverse Datasets for 1H NMR Serum Metabolic Profiling of Early Lactation Dairy Cows.

Most livestock metabolomic studies involve relatively small, homogenous populations of animals. However, livestock farming systems are non-homogenous, and large and more diverse datasets are required to ensure that biomarkers are robust. The aims of this study were therefore to (1) investigate the feasibility of using a large and diverse dataset for untargeted proton nuclear magnetic resonance (1H NMR) serum metabolomic profiling, and (2) investigate the impact of fixed effects (farm of origin, parity and stage of lactation) on the serum metabolome of early-lactation dairy cows. First, we used multiple linear regression to correct a large spectral dataset (707 cows from 13 farms) for fixed effects prior to multivariate statistical analysis with principal component analysis (PCA). Results showed that farm of origin accounted for up to 57% of overall spectral variation, and nearly 80% of variation for some individual metabolite concentrations. Parity and week of lactation had much smaller effects on both the spectra as a whole and individual metabolites (<3% and <20%, respectively). In order to assess the effect of fixed effects on prediction accuracy and biomarker discovery, we used orthogonal partial least squares (OPLS) regression to quantify the relationship between NMR spectra and concentrations of the current gold standard serum biomarker of energy balance, β-hydroxybutyrate (BHBA). Models constructed using data from multiple farms provided reasonably robust predictions of serum BHBA concentration (0.05 ≤ RMSE ≤ 0.18). Fixed effects influenced the results biomarker discovery; however, these impacts could be controlled using the proposed method of linear regression spectral correction.

Quantification of Analyte Concentration in the Single Molecule Regime Using Convolutional Neural Networks.

Single molecule (SM) detection represents the ultimate limit of chemical detection. Over the years, many experimental techniques have emerged with this capacity. Yet, SM detection and imaging methods produce large spectral data sets that benefit from chemometric methods. In particular, surface enhanced Raman scattering spectroscopy (SERS), with extensive applications in biosensing, is demonstrated to be particularly promising because Raman active molecules can be identified without recognition elements and is capable of SM detection. Yet quantification at ultralow analyte concentrations requiring detection of SM events remains an ongoing challenge, with the few existing methods requiring carefully developed calibration curves that must be redeveloped for each analyte molecule. In this work, we demonstrate that a convolutional neural network (CNN) model when applied to bundles of SERS spectra yields a robust, facile method for concentration quantification down to 10 fM using SM detection events. We further demonstrate that transfer learning, the process of reusing the weights of a trained CNN model, greatly reduces the amount of data required to train CNN models on new analyte molecules. These results point the way for unambiguous analysis of large spectral data sets and the use of SERS in important ultra low concentration chemical detection applications such as metabolomic profiling, water quality evaluation, and fundamental research.