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

Abstract

In this work, a novel probabilistic untargeted feature detection algorithm for liquid chromatography coupled to high-resolution mass spectrometry (LC-HRMS) using artificial neural network (ANN) is presented. The feature detection process is approached as a pattern recognition problem, and thus, ANN was utilized as an efficient feature recognition tool. Unlike most existing feature detection algorithms, with this approach, any suspected chromatographic profile (i.e., shape of a peak) can easily be incorporated by training the network, avoiding the need to perform computationally expensive regression methods with specific mathematical models. In addition, with this method, we have shown that the high-resolution raw data can be fully utilized without applying any arbitrary thresholds or data reduction, therefore improving the sensitivity of the method for compound identification purposes. Furthermore, opposed to existing deterministic (binary) approaches, this method rather estimates the probability of a feature being present/absent at a given point of interest, thus giving chance for all data points to be propagated down the data analysis pipeline, weighed with their probability. The algorithm was tested with data sets generated from spiked samples in forensic and food safety context and has shown promising results by detecting features for all compounds in a computationally reasonable time.

Self Adjusting Algorithm for the Nontargeted Feature Detection of High Resolution Mass Spectrometry Coupled with Liquid Chromatography Profile Data

“…Although there are numerous differences between the algorithms, they do share a common framework around the use of 2-dimensional data (i.e., centroided data , ) rather than 3-dimensional data (i.e., profile data) and the use of extracted ion chromatograms (e.g., XICs and/or region of interest , ). These approximations are made in order to reduce data size and consequently decrease the data processing time, but they come at the cost of the necessity for a suite of optimizable parameters that the users need to carefully set in order to minimize the rate of false detection. , However, multiple studies have shown that the feature detection using this procedure, even under optimized conditions, is prone to high rates of false detection. − As of today, there have been only a few studies working with the three-dimensional (3D) data. , One such method used a probabilistic approach, while the other one employs the artificial neural networks for the feature detection in the LC-HRMS data . The main disadvantages of these methods are the fact that they need to be trained and in the case of artificial neural networks the data needed to be binned prior to their use.…”

“…22−25 As of today, there have been only a few studies working with the three-dimensional (3D) data. 26,27 One such method used a probabilistic approach, 27 while the other one employs the artificial neural networks for the feature detection in the LC-HRMS data. 26 The main disadvantages of these methods are the fact that they need to be trained and in the case of artificial neural networks the data needed to be binned prior to their use.…”

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

“…The use of supervised machine learning algorithms, such as artificial neural networks (ANNs), for probabilistic feature recognition in LC-HRMS data has been suggested. Woldegebriel and Derks ( 2017 ) theorised that the detection of all possible peak features within a given sample can be considered a pattern recognition problem; therefore, a technique such as ANN can be especially useful. Features of interest within both the LC and MS space have unique characteristics, such as peak shapes and m/z patterns, that an algorithm can be trained to recognise.…”

“…The output of this algorithm provided a two-dimensional coordinate, including both RT and m/z , along with a posterior probability of whether these coordinates correspond to the centre of a peak feature. The authors noted that there was no correlation between the intensity of a signal and the probability of feature detection, indicating the ANN was generalising sufficiently and could identify all potential features within a sample (Woldegebriel and Derks 2017 ). The identified features could then be compared to libraries/databases or undergo further processing to achieve putative identification.…”

Spectroelectrochemical Sensing: Current Trends and Challenges

“…Therefore, it is important to develop mathematical tools to use all this information to solve complex analytical problems. Machine learning , neural networks , wavelets and many other mathematical tools will help to increase the potential applications of all the SEC techniques. Particularly, in the case of Raman‐SEC, users have to look for new mathematical tools to avoid the influence of the baseline in the spectral response.…”