Robust Bayesian Algorithm for Targeted Compound Screening in Forensic Toxicology

Abstract

As part of forensic toxicological investigation of cases involving unexpected death of an individual, targeted or untargeted xenobiotic screening of post-mortem samples is normally conducted. To this end, liquid chromatography (LC) coupled to high-resolution mass spectrometry (MS) is typically employed. For data analysis, almost all commonly applied algorithms are threshold-based (frequentist). These algorithms examine the value of a certain measurement (e.g., peak height) to decide whether a certain xenobiotic of interest (XOI) is present/absent, yielding a binary output. Frequentist methods pose a problem when several sources of information [e.g., shape of the chromatographic peak, isotopic distribution, estimated mass-to-charge ratio (m/z), adduct, etc.] need to be combined, requiring the approach to make arbitrary decisions at substep levels of data analysis. We hereby introduce a novel Bayesian probabilistic algorithm for toxicological screening. The method tackles the problem with a different strategy. It is not aimed at reaching a final conclusion regarding the presence of the XOI, but it estimates its probability. The algorithm effectively and efficiently combines all possible pieces of evidence from the chromatogram and calculates the posterior probability of the presence/absence of XOI features. This way, the model can accommodate more information by updating the probability if extra evidence is acquired. The final probabilistic result assists the end user to make a final decision with respect to the presence/absence of the xenobiotic. The Bayesian method was validated and found to perform better (in terms of false positives and false negatives) than the vendor-supplied software package.

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

“…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.…”

An artificial neural network system to identify alleles in reference electropherograms

“…A recent work by Taylor et al [1] demonstrated an artificial neural network (ANN) that was trained on two good quality reference EPGs to classify data in the 6-FAM dye lane and then applied to a third (also good quality) EPG with reasonable success. Taylor et al [1] provided a proof of concept that ANN could be used to interpret EPGs, which we extend here by: 1) Increasing the amount of training data 2) Increasing the range of training EPG quality from completely blank to highly overloaded 3) Improving on the architecture of the ANN used 4) Training a series of ANN that are used on different areas of the EPG 5) Coupling the predictions of the ANNs with a peak detection algorithm originally designed for LCMS data [2,3] and recently extended to DNA EPG data [4] to produce a peak detection and classification system Having created the peak detection and classification system we trial it on several profiles and demonstrate the results, which we compare to the peaks flagged by Genemapper® ID-X (Life Technologies).…”

Application of a Bayesian network to aid the interpretation of blood alcohol (ethanol) concentrations in air crashes

“…For example Fuller et al [10] presented a "causality index" to compare incidental and lethal post-mortem drug concentrations using inferential statistics , Langford et al [11] described a Bayesian Network for the assessment of the probability of a death having been caused by the analytically determined blood concentration of drug(s). Biedermann et al [12], Taroni et al [13]and Bossers & Paul [14] used likelihood ratios for the interpretation of forensic cut-offs and legal thresholds and Woldegebriel et al [15]used Bayesian algorithms for the detection of compounds during unknown drug screening.…”