Application of Fragment Ion Information as Further Evidence in Probabilistic Compound Screening Using Bayesian Statistics and Machine Learning: A Leap Toward Automation

Abstract

In this work, we introduce an automated, efficient, and elegant model to combine all pieces of evidence (e.g., expected retention times, peak shapes, isotope distributions, fragment-to-parent ratio) obtained from liquid chromatography-tandem mass spectrometry (LC-MS/MS/MS) data for screening purposes. Combining all these pieces of evidence requires a careful assessment of the uncertainties in the analytical system as well as all possible outcomes. To-date, the majority of the existing algorithms are highly dependent on user input parameters. Additionally, the screening process is tackled as a deterministic problem. In this work we present a Bayesian framework to deal with the combination of all these pieces of evidence. Contrary to conventional algorithms, the information is treated in a probabilistic way, and a final probability assessment of the presence/absence of a compound feature is computed. Additionally, all the necessary parameters except the chromatographic band broadening for the method are learned from the data in training and learning phase of the algorithm, avoiding the introduction of a large number of user-defined parameters. The proposed method was validated with a large data set and has shown improved sensitivity and specificity in comparison to a threshold-based commercial software package.

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

“…Unfortunately, most currently existing open-source/commercial software packages are not equipped to efficiently handle the large, high-resolution data sets, thus usually applying some sort of data reduction or preprocessing, vulnerable to discarding valuable information too early in the data analysis pipeline, resulting in a propagated error affecting the compound identification step. In our previous work ,, we had demonstrated that probabilistic peak detection is superior to deterministic algorithms and have shown that it outperforms widely used existing algorithms. In this work, a novel, fully untargeted, faster, and efficient probabilistic feature recognition algorithm based on artificial neural network (ANN) for LC–HRMS data has been developed, and its usefulness in forensic and food safety toxicology context has been demonstrated.…”

“…The mass spectrometer was operated in positive electrospray mode with an acquisition at 70 000 resolving power. Further details can be found on Materials and Methods section of refs and for the two data sets, respectively.…”

“…In the chromatographic domain, the elution process of a given compound retained by the column of the system is usually summarized as having a Gaussian-like shape with a long tail, and thus, it is a common approach to use a Gaussian function as a representative mathematical model for feature detection. − However, all factors affecting the peak shapes have not yet been fully characterized, so therefore, other mathematical models have also been proposed to represent chromatographic peaks . In most cases, the feature detection processes are approached either one dimensionally, by taking the chromatographic and MS space separately, or two-dimensionally, using three-dimensional models by taking into account both separation spaces simultaneously. , In our previous work, we have highlighted the benefit of using a multidimensional model to capture all common characteristics of an LC–HRMS data governed by a probabilistic framework. , In the second step of untargeted screening (ii), given the detected peak features from the first step (i), an exhaustive probabilistic assessment to identify a possible match between the detected features and any candidate compound that could explain the observed pattern should follow.…”

Machine Learning Supporting Experimental Design for Product Development in the Lab

“…Coming back to machine learning, it is worthwhile to mention that these techniques are increasingly applied to design of experiments, cf. . The goal is to build self‐learning, self‐monitoring and self‐optimizing chemical plants which plan and perform new experiments itself without using any further human resources.…”

Probabilistic peak detection in CE‐LIF for STR DNA typing

“…In this work, we introduce a novel probabilistic approach for peak detection in CE‐LIF epg's from STR DNA profiling data. Our Bayesian framework has been originally developed and introduced for chromatographic‐mass spectrometric data in analytical chemistry and proved to be ideally suited for this purpose . We theorize that the statistical evaluation of low‐level (complex) DNA profiles will be enhanced by including probabilistic peak detection data which are generated directly from CE data.…”