Artificial Neural Network for Probabilistic Feature Recognition in Liquid Chromatography Coupled to High-Resolution Mass Spectrometry

Deep learning framework for peak detection at the intact level of therapeutic proteins

While automated peak detection functionalities are available in commercially accessible software, achieving optimal true positive rates frequently necessitates visual inspection and manual adjustments. In the initial phase of this study, hetero-variants (glycoforms) of a monoclonal antibody were distinguished using liquid chromatography-mass spectrometry, revealing discernible peaks at the intact level. To comprehensively identify each peak (hetero-variant) in the intact-level analysis, a deep learning approach utilizing convolutional neural networks (CNNs) was employed in the subsequent phase of the study. In the current case study, utilizing conventional software for peak identification, five peaks were detected using a 0.5 threshold, whereas seven peaks were identified using the CNN model. The model exhibited strong performance with a probability area under the curve (AUC) of 0.9949, surpassing that of partial least squares discriminant analysis (PLS-DA) (probability AUC of 0.8041), and locally weighted regression (LWR) (probability AUC of 0.6885) on the data acquired during experimentation in real-time. The AUC of the receiver operating characteristic curve also illustrated the superior performance of the CNN over PLS-DA and LWR.

Applications of liquid chromatography-mass spectrometry based metabolomics in predictive and personalized medicine

Metabolomics is a fast-developing technique used in biomedical researches focusing on pathological mechanism illustration or novel biomarker development for diseases. The ability of simultaneously quantifying thousands of metabolites in samples makes metabolomics a promising technique in predictive or personalized medicine-oriented researches and applications. Liquid chromatography-mass spectrometry is the most widely employed analytical strategy for metabolomics. In this current mini-review, we provide a brief update on the recent developments and novel applications of LC-MS based metabolomics in the predictive and personalized medicine sector, such as early diagnosis, molecular phenotyping or prognostic evaluation. COVID-19 related metabolomic studies are also summarized. We also discuss the prospects of metabolomics in precision medicine-oriented researches, as well as critical issues that need to be addressed when employing metabolomic strategy in clinical applications.

Neural network in food analytics

Neural network (i.e. deep learning, NN)-based data analysis techniques have been listed as a pivotal opportunity to protect the integrity and safety of the global food supply chain and forecast $11.2 billion in agriculture markets. As a general-purpose data analytic tool, NN has been applied in several areas of food science, such as food recognition, food supply chain security and omics analysis, and so on. Therefore, given the rapid emergence of NN applications in food safety, this review aims to provide a comprehensive overview of the NN application in food analysis for the first time, focusing on domain-specific applications in food analysis by introducing fundamental methodology, reviewing recent and notable progress, and discussing challenges and potential pitfalls. NN demonstrated that it has a bright future through effective collaboration between food specialist and the broader community in the food field, for example, superiority in food recognition, sensory evaluation, pattern recognition of spectroscopy and chromatography. However, major challenges impeded NN extension including void in the food scientist-friendly interface software package, incomprehensible model behavior, multi-source heterogeneous data, and so on. The breakthrough from other fields proved NN has the potential to offer a revolution in the immediate future.

Developments in high-resolution mass spectrometric analyses of new psychoactive substances

The proliferation of new psychoactive substances (NPS) has necessitated the development and improvement of current practices for the detection and identification of known NPS and newly emerging derivatives. High-resolution mass spectrometry (HRMS) is quickly becoming the industry standard for these analyses due to its ability to be operated in data-independent acquisition (DIA) modes, allowing for the collection of large amounts of data and enabling retrospective data interrogation as new information becomes available. The increasing popularity of HRMS has also prompted the exploration of new ways to screen for NPS, including broad-spectrum wastewater analysis to identify usage trends in the community and metabolomic-based approaches to examine the effects of drugs of abuse on endogenous compounds. In this paper, the novel applications of HRMS techniques to the analysis of NPS is reviewed. In particular, the development of innovative data analysis and interpretation approaches is discussed, including the application of machine learning and molecular networking to toxicological analyses.

From Centroided to Profile Mode: Machine Learning for Prediction of Peak Width in HRMS Data

Centroiding is one of the major approaches used for size reduction of the data generated by high-resolution mass spectrometry. During centroiding, performed either during acquisition or as a pre-processing step, the mass profiles are represented by a single value (i.e., the centroid). While being effective in reducing the data size, centroiding also reduces the level of information density present in the mass peak profile. Moreover, each step of the centroiding process and their consequences on the final results may not be completely clear. Here, we present Cent2Prof, a package containing two algorithms that enables the conversion of the centroided data to mass peak profile data and vice versa. The centroiding algorithm uses the resolution-based mass peak width parameter as the first guess and self-adjusts to fit the data. In addition to the m/z values, the centroiding algorithm also generates the measured mass peak widths at half-height, which can be used during the feature detection and identification. The mass peak profile prediction algorithm employs a random-forest model for the prediction of mass peak widths, which is consequently used for mass profile reconstruction. The centroiding results were compared to the outputs of the MZmine-implemented centroiding algorithm. Our algorithm resulted in rates of false detection ≤5% while the MZmine algorithm resulted in 30% rate of false positive and 3% rate of false negative. The error in profile prediction was ≤56% independent of the mass, ionization mode, and intensity, which was 6 times more accurate than the resolution-based estimated values.

Machine learning meets omics: applications and perspectives

The innovation of biotechnologies has allowed the accumulation of omics data at an alarming rate, thus introducing the era of 'big data'. Extracting inherent valuable knowledge from various omics data remains a daunting problem in bioinformatics. Better solutions often need some kind of more innovative methods for efficient handlings and effective results. Recent advancements in integrated analysis and computational modeling of multi-omics data helped address such needs in an increasingly harmonious manner. The development and application of machine learning have largely advanced our insights into biology and biomedicine and greatly promoted the development of therapeutic strategies, especially for precision medicine. Here, we propose a comprehensive survey and discussion on what happened, is happening and will happen when machine learning meets omics. Specifically, we describe how artificial intelligence can be applied to omics studies and review recent advancements at the interface between machine learning and the ever-widest range of omics including genomics, transcriptomics, proteomics, metabolomics, radiomics, as well as those at the single-cell resolution. We also discuss and provide a synthesis of ideas, new insights, current challenges and perspectives of machine learning in omics.

Visual recognition and performance prediction of athletes based on target tracking EIA algorithm

In the past, the research of target tracking was often to track problems in a static background, and the tracking scenes were often stable, and the targets were special. However, target tracking is often a tracking problem in the face of realistic complex scenes, and the target and scene are more complex. Therefore, the target tracking algorithm still faces many challenges in practical applications, especially in sports visual feature recognition. Based on the needs of sports feature recognition, this study combines the EIA algorithm to construct a feature recognition model. Moreover, for the shortcomings of the compressed sensing tracking algorithm that cannot accurately and comprehensively describe the target shape through a single target feature, the multi-feature adaptive fusion method is used to visualize the target appearance model, thus improving the accuracy of target tracking. In addition, this study design experiments to analyze the performance of the algorithm model. The research results show that the algorithm model of this study has certain recognition effects.

High sensitivity and specificity feature detection in liquid chromatography-mass spectrometry data: A deep learning framework

Feature detection is a crucial pre-processing step for high-resolution liquid chromatography-mass spectrometry (LC-MS) data analysis. Typical practices based on thresholds or rigid mathematical assumptions can cause ineffective performance in detecting low abundance and non-ideal distributed compounds. We herein introduce a novel feature detection method based on deep learning named SeA-M2Net that considers feature detection as an image-based object detection task. By fully employing raw data directly, and integrating all related factors (e.g., LC elution, charge state, and isotope distribution) with two-dimensional pseudo color images to calculate the probability of the presence of the compound, low abundance compounds can be well preserved and observed. More importantly, SeA-M2Net, with deep multilevel and multiscale structures focuses on compound pattern detection in a learned method instead of assuming a mathematical parametric model. All parameters in SeA-M2Net are learned from data in the training procedure, thus allowing for maximum flexibility of pattern distribution deformation. The algorithm is tested on several LC-MS datasets of multiple biological samples obtained from different instruments with varied experimental settings. We demonstrate the superiority of the new approach in handling complex compound patterns (e.g., low abundance, overlapping regions, LC shifts, and missing values). Our experiments indicate that SeA-M2Net outperforms widely used detection methods in terms of detection accuracy.

Intelligent sports feature recognition system based on texture feature extraction and SVM parameter selection

Feature extraction is the basis of texture analysis. How to obtain texture features with small feature dimension, simple calculation and comprehensive representation of images is a hot spot and a difficult point in feature extraction. The traditional image texture feature extraction method is to process the image in the spatial domain. However, due to its high computational complexity, its practical application is restricted. Based on this, this study studies the extraction method of texture features, and deeply analyzes the principle of non-subsampled Contourlet transform. Moreover, this study uses NSCT to transform the image from the spatial domain to the frequency domain and extracts the texture features of the decomposed low frequency sub-band, intermediate frequency sub-band and high frequency sub-band image respectively. In addition, this study selects the appropriate parameters to establish the support vector machine model and applies the extracted texture features into the support vector machine for recognition and applies it to the sports feature recognition. Finally, this study designed a controlled experiment to analyze the performance of the algorithm. The results show that the proposed method has certain effects.

Research on athlete training behavior based on improved support vector algorithm and target image detection

In view of the defects and shortcomings of the traditional target detection and tracking algorithm in accurately detecting targets and targets in different scenarios, based on the current research status and technical level of target detection and tracking at home and abroad, this paper proposes a target detection algorithm and tracking method using neural network algorithm, and applies it to the athlete training model. Based on the Alex-Net network structure, this paper designs a three-layer convolutional layer and two layers of fully connected layers. The last layer is used as the input of the SVM classifier, and the target classification result is obtained by the SVM classifier. In addition, this article adds SPP-Layer between the convolutional layer and the fully connected layer, enabling the same dimension of the Feature Map to be obtained before the fully connected layer for different sized input images. The research results show that the proposed method has certain recognition effect and can be applied to athlete training.

Classification accuracy of basketball simulation training system based on sensor fusion and Bayesian algorithm

As a pattern recognition application direction, human body posture recognition provides decision-making basis for human body behavior pattern analysis of human-computer intelligent interactive control. Therefore, in a complete human-computer intelligent interaction system, human body posture recognition is a necessary link that can complete the human body’s behavioral characterization and make humanized decision-making. This paper studies the athlete’s posture recognition algorithm based on multi-sensor method and completes the whole process from data acquisition to data processing and model algorithm construction and verification. Moreover, this paper designs experiments to verify the model’s recognition results for athletes, and discusses the results, and analyzes the advantages and disadvantages of the model in this experiment. In addition, this study takes basketball action as an example to take identification analysis. The results show that the proposed method has certain practical effects and can provide theoretical reference for subsequent related research.

Research on sports retrieval recognition of action based on feature extraction and SVM classification algorithm

The feature extraction speed of the traditional athlete motion retrieval algorithm is slow, and it often takes dozens of minutes or even hours to analyze a video. The speed of this feature extraction obviously cannot meet the needs of big data video analysis. In response to these two problems exposed by Action Bank under large-scale data, this paper proposes to apply the template learning method based on spectral clustering to Action Bank, which replaces the cumbersome manual selection template step and is easy to generalize to different databases. Moreover, in view of the disadvantage of slow speed of extracting Action Bank features, this paper proposes a fast algorithm for accumulating Action Bank. In addition, this study uses the lookup table method instead of the time-consuming steps of the correlation distance calculation in template matching, which greatly accelerates the time of feature extraction. Finally, this study design experiments to analyze the performance of the algorithm. Through research, it can be seen that the algorithm of this study can be applied to athletes’ sports retrieval and has certain recognition effects.

Fake metabolomics chromatogram generation for facilitating deep learning of peak-picking neural networks

Finding peaks in chromatograms and determining their start and end points (peak picking) is a core task in chromatography based biotechnology. Construction of peak-picking neural networks by deep learning was, however, hampered from the preparation of exact peak-picked or "labeled" chromatograms since the exact start and end points were often unclear in overlapping peaks in real chromatograms. We present a design of a fake chromatogram generator, along with a method for deep learning of peak-picking neural networks. Fake chromatograms were generated by generation of fake peaks, random sampling of peak positions from feature distributions, and merging with real blank sample chromatograms. Information on the exact start and end points, as labeled on the fake chromatograms, were effective for training and evaluating peak-picking neural networks. The peak-picking neural networks constructed herein outperformed conventional peak-picking software and showed comparable performance with that of experienced operators for processing the widely targeted metabolome data. Results of this study indicate that generation of fake chromatograms would be crucial for developing peak-picking neural networks and a key technology for further improvement of peak picking neural networks.

Deep learning meets metabolomics: a methodological perspective

Deep learning (DL), an emerging area of investigation in the fields of machine learning and artificial intelligence, has markedly advanced over the past years. DL techniques are being applied to assist medical professionals and researchers in improving clinical diagnosis, disease prediction and drug discovery. It is expected that DL will help to provide actionable knowledge from a variety of 'big data', including metabolomics data. In this review, we discuss the applicability of DL to metabolomics, while presenting and discussing several examples from recent research. We emphasize the use of DL in tackling bottlenecks in metabolomics data acquisition, processing, metabolite identification, as well as in metabolic phenotyping and biomarker discovery. Finally, we discuss how DL is used in genome-scale metabolic modelling and in interpretation of metabolomics data. The DL-based approaches discussed here may assist computational biologists with the integration, prediction and drawing of statistical inference about biological outcomes, based on metabolomics data.

The application of artificial neural networks in metabolomics: a historical perspective

BACKGROUND

Metabolomics data, with its complex covariance structure, is typically modelled by projection-based machine learning (ML) methods such as partial least squares (PLS) regression, which project data into a latent structure. Biological data are often non-linear, so it is reasonable to hypothesize that metabolomics data may also have a non-linear latent structure, which in turn would be best modelled using non-linear equations. A non-linear ML method with a similar projection equation structure to PLS is artificial neural networks (ANNs). While ANNs were first applied to metabolic profiling data in the 1990s, the lack of community acceptance combined with limitations in computational capacity and the lack of volume of data for robust non-linear model optimisation inhibited their widespread use. Due to recent advances in computational power, modelling improvements, community acceptance, and the more demanding needs for data science, ANNs have made a recent resurgence in interest across research communities, including a small yet growing usage in metabolomics. As metabolomics experiments become more complex and start to be integrated with other omics data, there is potential for ANNs to become a viable alternative to linear projection methods.

AIM OF REVIEW

We aim to first describe ANNs and their structural equivalence to linear projection-based methods, including PLS regression. We then review the historical, current, and future uses of ANNs in the field of metabolomics.

KEY SCIENTIFIC CONCEPT OF REVIEW

Is metabolomics ready for the return of artificial neural networks?

Trace, Machine Learning of Signal Images for Trace-Sensitive Mass Spectrometry: A Case Study from Single-Cell Metabolomics

Recent developments in high-resolution mass spectrometry (HRMS) technology enabled ultrasensitive detection of proteins, peptides, and metabolites in limited amounts of samples, even single cells. However, extraction of trace-abundance signals from complex data sets ( m/ z value, separation time, signal abundance) that result from ultrasensitive studies requires improved data processing algorithms. To bridge this gap, we here developed "Trace", a software framework that incorporates machine learning (ML) to automate feature selection and optimization for the extraction of trace-level signals from HRMS data. The method was validated using primary (raw) and manually curated data sets from single-cell metabolomic studies of the South African clawed frog ( Xenopus laevis) embryo using capillary electrophoresis electrospray ionization HRMS. We demonstrated that Trace combines sensitivity, accuracy, and robustness with high data processing throughput to recognize signals, including those previously identified as metabolites in single-cell capillary electrophoresis HRMS measurements that we conducted over several months. These performance metrics combined with a compatibility with MS data in open-source (mzML) format make Trace an attractive software resource to facilitate data analysis for studies employing ultrasensitive high-resolution MS.