Deep learning-based component identification for the Raman spectra of mixtures

Raman spectrum matching with contrastive representation learning

An effective contrastive representation learning method for spectra identification with a frequentist guarantee of including the correct class prediction on two Raman datasets (Mineral and Organic) and one SERS dataset (Bacteria).

Dynamic one-shot target detection and classification using a pseudo-Siamese network and its application to Raman spectroscopy

Target detection and classification by Raman spectroscopy are important techniques for biological and chemical defense in military operations. Conventionally, these techniques preprocess the observed spectra using smoothing or baseline correction and apply detection algorithms like the generalized likelihood ratio test, independent component analysis, nonnegative matrix factorization, etc. These conventional detection algorithms need preprocessing and multiple shots of Raman spectra to get a reasonable accuracy. Recently, techniques based on deep learning are being used for target detection and classification due to its great adaptability and high accuracy over other methods and due to no requirement for preprocessing. Deep learning may give a good performance, but need retraining when untrained class targets are introduced which is time-consuming and bothersome. We devise a novel algorithm using a variant of the pseudo-Siamese network, one of the deep learning algorithms, that does not need retraining to detect and classify untrained class targets. Our algorithm detects and classifies targets with only one shot. In addition, our algorithm does not need preprocessing. We verify our algorithm with Raman spectra measured using a Raman spectrometer.

Retention time prediction in hydrophilic interaction liquid chromatography with graph neural network and transfer learning

The combination of retention time (RT), accurate mass and tandem mass spectra can improve the structural annotation in untargeted metabolomics. However, the incorporation of RT for metabolite identification has received less attention because of the limitation of available RT data, especially for hydrophilic interaction liquid chromatography (HILIC). Here, the Graph Neural Network-based Transfer Learning (GNN-TL) is proposed to train a model for HILIC RTs prediction. The graph neural network was pre-trained using an in silico HILIC RT dataset (pseudo-labeling dataset) with ∼306 K molecules. Then, the weights of dense layers in the pre-trained GNN (pre-GNN) model were fine-tuned by transfer learning using a small number of experimental HILIC RTs from the target chromatographic system. The GNN-TL outperformed the methods in Retip, including the Random Forest (RF), Bayesian-regularized neural network (BRNN), XGBoost, light gradient-boosting machine (LightGBM), and Keras. It achieved the lowest mean absolute error (MAE) of 38.6 s on the test set and 33.4 s on an additional test set. It has the best ability to generalize with a small performance difference between training, test, and additional test sets. Furthermore, the predicted RTs can filter out nearly 60% false positive candidates on average, which is valuable for the identification of compounds complementary to mass spectrometry.

Detection of pancreatic cancer by convolutional-neural-network-assisted spontaneous Raman spectroscopy with critical feature visualization

Pancreatic cancer is the deadliest cancer type with a five-year survival rate of less than 9%. Detection of tumor margins plays an essential role in the success of surgical resection. However, histopathological assessment is time-consuming, expensive, and labor-intensive. We constructed a lab-designed, hand-held Raman spectroscopic system that could enable intraoperative tissue diagnosis using convolutional neural network (CNN) models to efficiently distinguish between cancerous and normal pancreatic tissue. To our best knowledge, this is the first reported effort to diagnose pancreatic cancer by CNN-aided spontaneous Raman scattering with a lab-developed system designed for intraoperative applications. Classification based on the original one-dimensional (1D) Raman, two-dimensional (2D) Raman images, and the first principal component (PC1) from the principal component analysis on the 2D image, could all achieve high performance: the testing sensitivity, specificity, and accuracy were over 95%, and the area under the curve approached 0.99. Although CNN models often show great success in classification, it has always been challenging to visualize the CNN features in these models, which has never been achieved in the Raman spectroscopy application in cancer diagnosis. By studying individual Raman regions and by extracting and visualizing CNN features from max-pooling layers, we identified critical Raman peaks that could aid in the classification of cancerous and noncancerous tissues. 2D Raman PC1 yielded more critical peaks for pancreatic cancer identification than that of 1D Raman, as the Raman intensity was amplified by 2D Raman PC1. To our best knowledge, the feature visualization was achieved for the first time in the field of CNN-aided spontaneous Raman spectroscopy for cancer diagnosis. Based on these CNN feature peaks and their frequency at specific wavenumbers, pancreatic cancerous tissue was found to contain more biochemical components related to the protein contents (particularly collagen), whereas normal pancreatic tissue was found to contain more lipids and nucleic acid (particularly deoxyribonucleic acid/ribonucleic acid). Overall, the CNN model in combination with Raman spectroscopy could serve as a useful tool for the extraction of key features that can help differentiate pancreatic cancer from a normal pancreas.

Exploring machine learning methods for absolute configuration determination with vibrational circular dichroism

The added value of supervised Machine Learning (ML) methods to determine the Absolute Configuration (AC) of compounds from their Vibrational Circular Dichroism (VCD) spectra was explored. Among all ML methods considered, Random Forest (RF) and Feedforward Neural Network (FNN) yield the best performance for identification of the AC. At its best, FNN allows near-perfect AC determination, with accuracy of prediction up to 0.995, while RF combines good predictive accuracy (up to 0.940) with the ability to identify the spectral areas important for the identification of the AC. No loss in performance of either model is observed as long as the spectral sampling interval used does not exceed the spectral bandwidth. Increasing the sampling interval proves to be the best method to lower the dimensionality of the input data, thereby decreasing the computational cost associated with the training of the models.

Field determination of hazardous chemicals in public security by using a hand-held Raman spectrometer and a deep architecture-search network

With the advanced development of miniaturization and integration of instruments, Raman spectroscopy (RS) has demonstrated its great significance because of its non-invasive property and fingerprint identification ability, and extended its applications in public security, especially for hazardous chemicals. However, the fast and accurate RS analysis of hazardous chemicals in field test by non-professionals is still challenging due to the lack of an effective and timely spectral-based chemical-discriminating solution. In this study, a platform was developed for the field determination of hazardous chemicals in public security by using a hand-held Raman spectrometer and a deep architecture-search network (DASN) incorporated into a cloud server. With the Raman spectra of 300 chemicals, DASN stands out with identification accuracy of 100% and outweighs other machine learning and deep learning methods. The network feature maps for the spectra of methamphetamine and ketamine focus on the main peaks of 1001 and 652 cm^-1, which indicates the powerful feature extraction capability of DASN. Its receiver operating characteristic (ROC) curve completely encloses the other models, and the area under the curve is up to 1, implying excellent robustness. With the well-built platform combining RS, DASN, and cloud server, one test process including Raman measurement and identification can be performed in tens of seconds. Hence, the developed platform is simple, fast, accurate, and could be considered as a promising tool for hazardous chemical identification in public security on the scene.

Robustness and accuracy improvement of data processing with 2D neural networks for transient absorption dynamics

To achieve the goal of efficiently analyzing transient absorption spectra without arbitrary assumption and to overcome the limitations of conventional methods in fitting ability and highly noised backgrounds, it is essential to develop new tools to achieve more accurate and robust prediction based on the intrinsic properties of a spectrum even under strong noise. In this work, Lasso regression and neural network were combined to achieve an effective fitting. Compared to the conventional global fitting method, our network could automatically determine the exponential form on each wave unit, in which the accuracy was as high as 97%. Thereafter, the lifetime with the corresponding amplitude ratio could be easily predicted by the neural network on each wave unit. This kind of prediction is difficult to achieve by global fitting due to the limitation of computational resources. Furthermore, more accurate fitting even under weak signals could be achieved for the mean square error (MSE) decreasing by more than 100 times on average compared to conventional global fitting methods. Attributed to its improved accuracy and robustness, our developed algorithm could be readily applied to analyze time-resolved transient spectra.

Analysis of Raman Spectra by Using Deep Learning Methods in the Identification of Marine Pathogens

The need for efficient and accurate identification of pathogens in seafood and the environment has become increasingly urgent, given the current global pandemic. Traditional methods are not only time consuming but also lead to sample wastage. Here, we have proposed two new methods that involve Raman spectroscopy combined with a long short-term memory (LSTM) neural network and compared them with a method using a normal convolutional neural network (CNN). We used eight strains isolated from the marine organism Urechis unicinctus, including four kinds of pathogens. After the models were configured and trained, the LSTM methods that we proposed achieved average isolation-level accuracies exceeding 94%, not only meeting the requirement for identification but also indicating that the proposed methods were faster and more accurate than the normal CNN models. Finally, through a computational approach, we designed a loss function to explore the mechanism reflected by the Raman data, finding the Raman segments that most likely exhibited the characteristics of nucleic acids. These novel experimental results provide insights for developing additional deep learning methods to accurately analyze complex Raman data.

Development of Crime Scene Intelligence Using a Hand-Held Raman Spectrometer and Transfer Learning

The classification of ignitable liquids, such as gasoline, is critical crime scene intelligence to assist arson investigations. Rapid field gasoline classification is challenging because the current forensic testing standard requires gas chromatography-mass spectrometry analysis of evidence in an accredited laboratory. In this work, we reported a new intelligent analytical platform for field identification and classification of gasoline evidence. A hand-held Raman spectrometer was utilized to collect Raman spectra of reference gasoline samples with various octane numbers. The Raman spectrum pattern was converted into image presentations by continuous wavelet transformation (CWT) to facilitate artificial intelligence development using the transfer learning technique. GoogLeNet, a pretrained convolutional neural network (CNN), was adapted to train the classification model. Six different classification models were also developed from the same data set using conventional machine learning algorithms to evaluate the performance of our new approach. The experimental results indicated that the pretrained CNN model developed by our new data workflow outperformed other models in several performance benchmarks, such as accuracy, precision, recall, F1, Cohen's Kappa, and Matthews correlation coefficient. When the transfer learning model was challenged with the data collected from weathered gasoline samples, the classifier could still offer 73 and 53% accuracy for 50 and 25% weathered gasoline samples, respectively. In conclusion, wavelet transforms combined with transfer learning successfully processed and classified complex Raman spectral data without feature engineering. We envision that this nondestructive, automated, and accurate platform will accelerate crime scene intelligence development based on evidence's chemical signatures detected by hand-held Raman spectrometers.

Determination of the Geographical Origin of Coffee Beans Using Terahertz Spectroscopy Combined With Machine Learning Methods

Different geographical origins can lead to great variance in coffee quality, taste, and commercial value. Hence, controlling the authenticity of the origin of coffee beans is of great importance for producers and consumers worldwide. In this study, terahertz (THz) spectroscopy, combined with machine learning methods, was investigated as a fast and non-destructive method to classify the geographic origin of coffee beans, comparing it with the popular machine learning methods, including convolutional neural network (CNN), linear discriminant analysis (LDA), and support vector machine (SVM) to obtain the best model. The curse of dimensionality will cause some classification methods which are struggling to train effective models. Thus, principal component analysis (PCA) and genetic algorithm (GA) were applied for LDA and SVM to create a smaller set of features. The first nine principal components (PCs) with an accumulative contribution rate of 99.9% extracted by PCA and 21 variables selected by GA were the inputs of LDA and SVM models. The results demonstrate that the excellent classification (accuracy was 90% in a prediction set) could be achieved using a CNN method. The results also indicate variable selecting as an important step to create an accurate and robust discrimination model. The performances of LDA and SVM algorithms could be improved with spectral features extracted by PCA and GA. The GA-SVM has achieved 75% accuracy in a prediction set, while the SVM and PCA-SVM have achieved 50 and 65% accuracy, respectively. These results demonstrate that THz spectroscopy, together with machine learning methods, is an effective and satisfactory approach for classifying geographical origins of coffee beans, suggesting the techniques to tap the potential application of deep learning in the authenticity of agricultural products while expanding the application of THz spectroscopy.

Interpretable deep learning-assisted laser-induced breakdown spectroscopy for brand classification of iron ores

Brand classification of iron ores using laser-induced breakdown spectroscopy (LIBS) combined with artificial neural networks can quickly realize the compliance verification and guarantee the interests of both trading partners. However, its practical application is impeded by complex pretreatments and unexplained feature learning problems. According to the LIBS data characteristics of iron ores, a convolutional neural network (CNN) is designed to predict 16 types of brand iron ores from Australia, Brazil, and South Africa. The accuracies of the calibration set and the prediction set with five-fold cross-validation (5-CV) were 99.86% and 99.88%, and the value of loss function was 0.0356. Meanwhile, the established CNN method was also compared with common machine learning methods using raw spectra as input variables, and it outperformed other methods. For the first time, this work interprets the CNN's effectiveness layer by layer in self-adaptively extracting LIBS features through t-distributed stochastic neighbor embedding (t-SNE) and the quantitative data of major chemical components in iron ores. Our approach shows that deep learning assisted LIBS is able to significantly reduce manual factors in preprocessing and feature selection and has broad application prospects in the brand classification of iron ores.

Deep Learning for Biospectroscopy and Biospectral Imaging: State-of-the-Art and Perspectives

With the advances in instrumentation and sampling techniques, there is an explosive growth of data from molecular and cellular samples. The call to extract more information from the large data sets has greatly challenged the conventional chemometrics method. Deep learning, which utilizes very large data sets for finding hidden features therein and for making accurate predictions for a wide range of applications, has been applied in an unbelievable pace in biospectroscopy and biospectral imaging in the recent 3 years. In this Feature, we first introduce the background and basic knowledge of deep learning. We then focus on the emerging applications of deep learning in the data preprocessing, feature detection, and modeling of the biological samples for spectral analysis and spectroscopic imaging. Finally, we highlight the challenges and limitations in deep learning and the outlook for future directions.

Rapid on-site identification of pesticide residues in tea by one-dimensional convolutional neural network coupled with surface-enhanced Raman scattering

In this study, a novel analytical approach is proposed for the identification of pesticide residues in tea by combining surface-enhanced Raman scattering (SERS) with a deep learning method one-dimensional convolutional neural network (1D CNN). First, a handheld Raman spectrometer was used for rapid on-site collection of SERS spectra. Second, the collected SERS spectra were augmented by a data augmentation strategy. Third, based on the augmented SERS spectra, the 1D CNN models were established on the cloud server, and then the trained 1D CNN models were used for subsequent pesticide residue identification analysis. In addition, to investigate the identification performance of the 1D CNN method, four conventional identification methods, including partial least square-discriminant analysis (PLS-DA), k-nearest neighbour (k-NN), support vector machine (SVM) and random forest (RF), were also developed on the basis of the augmented SERS spectra and applied for pesticide residue identification analysis. The comparative studies show that the 1D CNN method possesses better identification accuracy, stability and sensitivity than the other four conventional identification methods. In conclusion, the proposed novel analytical approach that exploits the advantages of SERS and a deep learning method (1D CNN) is a promising method for rapid on-site identification of pesticide residues in tea.

Potential of Raman spectroscopy in facilitating pharmaceutical formulations development - An AI perspective

Drug development is time-consuming and inherently possesses a high failure rate. Pharmaceutical formulation development is the bridge that links a new chemical entity (NCE) to pre-clinical and clinical trials, and has a high impact on the efficacy and safety of the final drug product. Further, the time required for this process is escalating as formulation techniques are becoming more complicated due to the rising demands for drug products with better efficacy and patient compliance, as well as the inherent difficulties of addressing the unfavorable properties of NCEs such as low water solubility. The advent of artificial intelligence (AI) provides possibilities to accelerate the drug development process. In this review, we first examine applications of AI methods in different types of pharmaceutical formulations and formulation techniques. Moreover, as availability of data is the engine for the advancement of AI, we then suggest a potential way (i.e. applying Raman spectroscopy) for faster high-quality data gathering from formulations. Raman techniques have the capability of analyzing the composition and distribution of components and the physicochemical properties thereof within formulations, which are prominent factors governing drug dissolution profiles and subsequently bioavailability. Thus, useful information can be obtained bridging formulation development to the final product quality.

Quantitative analysis of excipient dominated drug formulations by Raman spectroscopy combined with deep learning

Owing to the growing interest in the application of Raman spectroscopy for quantitative purposes in solid pharmaceutical preparations, an article on the identification of compositions in excipient dominated drugs based on Raman spectra is presented. We proposed label-free Raman spectroscopy in conjunction with deep learning (DL) and non-negative least squares (NNLS) as a solution to overcome the drug fast screening bottleneck, which is not only a great challenge to drug administration, but also a major scientific challenge linked to falsified and/or substandard medicines. The result showed that Raman spectroscopy remains a cost effective, rapid, and user-friendly method, which if combined with DL and NNLS leads to fast implantation in the identification of lactose dominated drug (LDD) formulations. Meanwhile, Raman spectroscopy with the peak matching method allows a visual interpretation of the spectral signature (presence or absence of active pharmaceutical ingredients (APIs) and low content APIs).

Deep metabolome: Applications of deep learning in metabolomics

In the past few years, deep learning has been successfully applied to various omics data. However, the applications of deep learning in metabolomics are still relatively low compared to others omics. Currently, data pre-processing using convolutional neural network architecture appears to benefit the most from deep learning. Compound/structure identification and quantification using artificial neural network/deep learning performed relatively better than traditional machine learning techniques, whereas only marginally better results are observed in biological interpretations. Before deep learning can be effectively applied to metabolomics, several challenges should be addressed, including metabolome-specific deep learning architectures, dimensionality problems, and model evaluation regimes.

Classification of pathogens by Raman spectroscopy combined with generative adversarial networks

Rapid identification of marine pathogens is very important in marine ecology. Artificial intelligence combined with Raman spectroscopy is a promising choice for identifying marine pathogens due to its rapidity and efficiency. However, considering the cost of sample collection and the challenging nature of the experimental environment, only limited spectra are typically available to build a classification model, which hinders qualitative analysis. In this paper, we propose a novel method to classify marine pathogens by means of Raman spectroscopy combined with generative adversarial networks (GANs). Three marine strains, namely, Staphylococcus hominis, Vibrio alginolyticus, and Bacillus licheniformis, were cultured. Using Raman spectroscopy, we acquired 100 spectra of each strain, and we fitted them into GAN models for training. After 30,000 training iterations, the spectra generated by G were similar to the actual spectra, and D was used to test the accuracy of the spectra. Our results demonstrate that our method not only improves the accuracy of machine learning classification but also solves the problem of requiring a large amount of training data. Moreover, we have attempted to find potential identifying regions in the Raman spectra that can be used for reference in subsequent related work in this field. Therefore, this method has tremendous potential to be developed as a tool for pathogen identification.

Rise of Raman spectroscopy in neurosurgery: a review

SIGNIFICANCE

Although the clinical potential for Raman spectroscopy (RS) has been anticipated for decades, it has only recently been used in neurosurgery. Still, few devices have succeeded in making their way into the operating room. With recent technological advancements, however, vibrational sensing is poised to be a revolutionary tool for neurosurgeons.

AIM

We give a summary of neurosurgical workflows and key translational milestones of RS in clinical use and provide the optics and data science background required to implement such devices.

APPROACH

We performed an extensive review of the literature, with a specific emphasis on research that aims to build Raman systems suited for a neurosurgical setting.

RESULTS

The main translatable interest in Raman sensing rests in its capacity to yield label-free molecular information from tissue intraoperatively. Systems that have proven usable in the clinical setting are ergonomic, have a short integration time, and can acquire high-quality signal even in suboptimal conditions. Moreover, because of the complex microenvironment of brain tissue, data analysis is now recognized as a critical step in achieving high performance Raman-based sensing.

CONCLUSIONS

The next generation of Raman-based devices are making their way into operating rooms and their clinical translation requires close collaboration between physicians, engineers, and data scientists.

Determination of causes of death via spectrochemical analysis of forensic autopsies-based pulmonary edema fluid samples with deep learning algorithm

This study investigated whether infrared spectroscopy combined with a deep learning algorithm could be a useful tool for determining causes of death by analyzing pulmonary edema fluid from forensic autopsies. A newly designed convolutional neural network-based deep learning framework, named DeepIR and eight popular machine learning algorithms, were used to construct classifiers. The prediction performances of these classifiers demonstrated that DeepIR outperformed the machine learning algorithms in establishing classifiers to determine the causes of death. Moreover, DeepIR was generally less dependent on preprocessing procedures than were the machine learning algorithms; it provided the validation accuracy with a narrow range from 0.9661 to 0.9856 and the test accuracy ranging from 0.8774 to 0.9167 on the raw pulmonary edema fluid spectral dataset and the nine preprocessing protocol-based datasets in our study. In conclusion, this study demonstrates that the deep learning-equipped Fourier transform infrared spectroscopy technique has the potential to be an effective aid for determining causes of death.

Deep-learning-enhanced ice thickness measurement using Raman scattering

In ice thickness measurement (ICM) procedures based on Raman scattering, a key issue is the detection of ice-water interface using the slight difference between the Raman spectra of ice and water. To tackle this issue, we developed a new deep residual network (DRN) to cast this detection as an identification problem. Thus, the interface detection is converted to the prediction of the Raman spectra of ice and water. We enabled this process by designing a powerful DRN that was trained by a set of Raman spectral data, obtained in advance. In contrast to the state-of-the-art Gaussian fitting method (GFM), the proposed DRN enables ICM with a simple operation and low costs, as well as high accuracy and speed. Experimental results were collected to demonstrate the feasibility and effectiveness of the proposed DRN.

Machine learning based temperature prediction of poly(N-isopropylacrylamide)-capped plasmonic nanoparticle solutions

The temperature-dependent optical properties of gold nanoparticles that are capped with the thermo-sensitive polymer: 'poly(N-isopropylacrylamide)' (PNIPAM), have been studied extensively for several years. Also, their suitability to function as nanoscopic thermometers for bio-sensing applications has been suggested numerous times. In an attempt to establish this, many have studied the temperature-dependent optical resonance characteristics of these particles; however, developing a simple mathematical relationship between the optical measurements and the solution temperature remains an open challenge. In this paper, we attempt to systematically address this problem using machine learning techniques to quickly and accurately predict the solution-temperature, based on spectroscopic data. Our emphasis is on establishing a simple and practically useful solution to this problem. Our dataset comprises spectroscopic absorption data from both nanorods and nanobipyramids capped with PNIPAM, measured at discretely varied and pre-set temperature states. Specific regions of the spectroscopic data are selected as features for prediction using random forest (RF), gradient boosting (GB) and adaptive boosting (AB) regression techniques. Our prediction results indicate that RF and GB techniques can be used successfully to predict solution temperatures instantly to within 1 °C of accuracy.

A Toolbox for Organelle Mechanobiology Research-Current Needs and Challenges

Mechanobiology studies from the last decades have brought significant insights into many domains of biological research, from development to cellular signaling. However, mechano-regulation of subcellular components, especially membranous organelles, are only beginning to be unraveled. In this paper, we take mitochondrial mechanobiology as an example to discuss recent advances and current technical challenges in this field. In addition, we discuss the needs for future toolbox development for mechanobiological research of intracellular organelles.

Fast and accurate decoding of Raman spectra-encoded suspension arrays using deep learning

A deep learning network called “residual neural network” (ResNet) was used to decode Raman spectra-encoded suspension arrays (SAs).