Development of an Enhanced Total Ion Current Chromatogram Algorithm to Improve Untargeted Peak Detection

Enhancing gas chromatography-mass spectrometry resolution and pure analyte discovery using intra-chromatogram elution profile matching

Chemometric decomposition methods like multivariate curve resolution-alternating least squares (MCR-ALS) are often employed in gas chromatography-mass spectrometry (GC-MS) to improve analyte identification and quantitation. However, these methods can perform poorly for analytes with a low chromatographic resolution (Rs) and a high degree of spectral contamination from noise and background interferences. Thus, we propose a novel computational algorithm, termed mzCompare, to improve analyte identification and quantitation when coupled to MCR-ALS. The mzCompare method utilizes an underlying requirement that the retention time and peak shape between mass channels (m/z) of the same analyte should be similar. By discovering the selective m/z for a given analyte in a chromatogram, a pure elution profile can be generated and used as an equality constraint in MCR-ALS. The performance of the mzCompare methodology is demonstrated with both experimental and simulated chromatograms. Experimentally, unresolved analytes with a Rs as low as 0.05 could be confidently identified with mzCompare assisted MCR-ALS. Furthermore, application of the mzCompare algorithm to a complex aerospace fuel resulted in the discovery of 335 analytes, a 44 % increase compared to conventional peak detection methods. GC-MS simulations of target-interferent analyte pairs demonstrated that the performance of MCR-ALS deteriorated below a Rs of ∼0.25. However, mzCompare assisted MCR-ALS showed excellent identification and acceptable quantitative accuracy at a Rs of ∼0.02. These results show that the mzCompare algorithm can help analysts overcome modeling ambiguities resulting from the chemometric multiplex disadvantage.

Highly automatic and universal approach for pure ion chromatogram construction from liquid chromatography-mass spectrometry data using deep learning

Feature extraction is the most fundamental step when analyzing liquid chromatography-mass spectrometry (LC-MS) datasets. However, traditional methods require optimal parameter selections and re-optimization for different datasets, thus hindering efficient and objective large-scale data analysis. Pure ion chromatogram (PIC) is widely used because it avoids the peak splitting problem of the extracted ion chromatogram (EIC) and regions of interest (ROIs). Here, we developed a deep learning-based pure ion chromatogram method (DeepPIC) to find PICs using a customized U-Net from centroid mode data of LC-MS directly and automatically. A model was trained, validated, and tested on the Arabidopsis thaliana dataset with 200 input-label pairs. DeepPIC was integrated into KPIC2. The combination enables the entire processing pipeline from raw data to discriminant models for metabolomics datasets. The KPIC2 with DeepPIC was compared against other competing methods (XCMS, FeatureFinderMetabo, and peakonly) on the MM48, simulated MM48, and quantitative datasets. These comparisons showed that DeepPIC outperforms XCMS, FeatureFinderMetabo, and peakonly in recall rates and correlation with sample concentrations. Five datasets of different instruments and samples were used to evaluate the quality of PICs and the universal applicability of DeepPIC, and 95.12% of the found PICs could precisely match their manually labeled PICs. Therefore, KPIC2+DeepPIC is an automatic, practical, and off-the-shelf method to extract features from raw data directly, exceeding traditional methods with careful parameter tuning. It is publicly available at https://github.com/yuxuanliao/DeepPIC.

A fast stop-flow two-dimensional liquid chromatography tandem mass spectrometry and its application in food-derived protein hydrolysates

Food-derived bioactive peptides have many outstanding features like high safety, easy absorption, etc. However, explorations of the peptides are suffering from the limited knowledge of sample composition and low efficiency of separation techniques. In this work, a fast stop-flow two-dimensional liquid chromatography tandem mass spectrometry (2DLC-MS) was designed and constructed in-house. For chromatographic system optimization, the effects of column pairs and fraction transfer volumes on separation performance were studied. The pair of Protein BEH SEC and HSS T3 columns was found of high orthogonality. The peak capacity detected by the optimized 2DLC reached 1165 (for corn protein hydrolysates), indicating high resolving power. Moreover, the number of peptides identified from corn, soybean and casein protein hydrolysates reached as high as 8330, 8925 and 7215, respectively, demonstrating the high potential of the system. This would help reveal the peptide composition and facilitate the research on exploring bioactive peptides from food-derived protein hydrolysates.

Investigating analyte breakthrough under non-linear isotherm conditions during solid phase extraction facilitated by non-targeted analysis with comprehensive two-dimensional gas chromatography time-of-flight mass spectrometry

Solid phase extraction (SPE) sample preparation for the analysis of complex organic mixtures is often applied assuming all analytes of interest will preconcentrate on the stationary phase. This assumption ignores the reality that extraction is a dynamic interactive process and a diverse range of affinities for the stationary phase will result in equally diverse breakthrough volumes due to competitive sorption processes. To study this dynamic interactive process, and further to take advantage of it, we extracted a JP-8 jet fuel spiked with 40 ppm of a polar compound mix with silica and alumina SPE cartridges and analyzed sequential extracted fractions of the fuel to both assess the shifting chemical landscape present in the extraction and the impact of both SPE stationary phases on this process. Tile-based 1v1 comparative analysis (a recently reported extension of tile-based Fisher ratio analysis) was used to discover the (polar) compounds whose concentrations change between extracted fractions, discovering 21 compounds extracted with silica and 27 compounds extracted with alumina with at least a 2-fold change in concentration from the neat sample relative to the first 1 mL pass fraction sample. These compounds were quantified in each fraction to construct concentration ratio profiles, defined as the concentration ratio for a given SPE fraction per analyte compound relative to the analyte concentration in the neat fuel, for which the extraction behavior for each analyte could be assessed. These analyte compounds were found to breakthrough at different rates, with some analytes remaining on the column indefinitely (until extracted with a subsequent polar solvent) and other analytes eluting before the extraction is complete. Furthermore, in a comparison of the effect of selected stationary phase, alumina was found to retain oxygen-containing phenolic compounds to a greater extent than silica. Principal component analysis (PCA) was used to analyze the concentration ratio profiles of the various trace analytes in the JP8 fuel (phenols, indoles, etc.) in the context of their stationary phase affinity (silica or alumina) and competitive sorption behavior.

Enhancing partial least squares modeling of comprehensive two-dimensional gas chromatography time-of-flight mass spectrometry data by tile-based variance ranking

Chemometric methods like partial least squares (PLS) regression are valuable for correlating sample-based differences hidden in comprehensive two-dimensional gas chromatography (GC × GC) data to independently measured physicochemical properties. Herein, this work establishes the first implementation of tile-based variance ranking as a selective data reduction methodology to improve PLS modeling performance of 58 diverse aerospace fuels. Tile-based variance ranking discovered a total of 521 analytes with a square of the relative standard deviation (RSD²) in signal between 0.07 to 22.84. The goodness-of-fit for the models were determined by their normalized root-mean-square error of cross-validation (NRMSECV) and normalized root-mean-square error of prediction (NRMSEP). PLS models developed for viscosity, hydrogen content, and heat of combustion using all 521 features discovered by tile-based variance ranking had a respective NRMSECV (NRMSEP) equal to 10.5 % (10.2 %), 8.3 % (7.6 %), and 13.1 % (13.5 %). In contrast, use of a single-grid binning scheme, a common data reduction strategy for PLS analysis, resulted in less accurate models for viscosity (NRMSECV = 14.2 %; NRMSEP = 14.3 %), hydrogen content (NRMSECV = 12.1 %; NRMSEP = 11.0 %), and heat of combustion (NRMSECV = 14.4 %; NRMSEP = 13.6 %). Further, the features discovered by tile-based variance ranking can be optimized for each PLS model with RReliefF analysis, a machine learning algorithm. RReliefF feature optimization selected 48, 125, and 172 analytes out of the original 521 discovered by tile-based variance ranking to model viscosity, hydrogen content, and heat of combustion, respectively. The RReliefF optimized features developed highly accurate property-composition models for viscosity (NRMSECV = 7.9 %; NRMSEP = 5.8 %), hydrogen content (NRMSECV = 7.0 %; NRMSEP = 4.9 %), heat of combustion (NRMSECV = 7.9 %; NRMSEP = 8.4 %). This work also demonstrates that processing the chromatograms with a tile-based approach allows the analyst to directly identify the analytes of importance in a PLS model. Coupling tile-based feature selection with PLS analysis allows for deeper understanding in any property-composition study.

Optimizing column-to-column retention time alignment in high-speed gas chromatography by combining retention time locking and correlation optimized warping

We examine and then optimize alignment of chromatograms collected on nominally identical columns using retention time locking (RTL), an instrumental alignment tool, and software-based alignment using correlation optimized warping (COW). For this purpose, three samples are constructed by spiking two sets of analytes into a base test mixture. The three samples are analyzed by high-speed gas chromatography with four nominally identical columns and identical separation conditions. The data is first analyzed without alignment, then using COW alone, then RTL alone, and finally with RTL followed by COW to correct the severe column-to-column misalignment. Principal component analysis (PCA) is used to investigate how well each alignment method clustered the chromatograms into the three sample classes via a scores plot without being compromised by the specific column(s) used. The degree-of-class separation (DCS) is used as a classification metric, measured as the Euclidian distance between the centroids of two clusters in PC space in the scores plot, normalized by their pooled variance. With no alignment, the average DCS between sample classes (DCS_sam) was 3.0, while the average DCS between the four nominally identical columns, i.e., column classes (DCS_col) was 76.1 (ideally the DCS_col should be 0), indicating the chromatograms were initially classified by the columns used. Using either COW or RTL alone also produced unsatisfactory results, with COW alone incorrectly aligning many peaks, leading to a DCS_sam of only 1.9 and DCS_col of 1.7, while RTL alone provided a DCS_sam of 4.7 and DCS_col of 4.2. Finally, using RTL followed by COW alignment, DCS_sam increased to 32.5, indicating successful classification by chemical differences between sample classes, while the DCS_col decreased to 0.4, indicating virtually no classification due to column-to-column differences, as desired. Thus, RTL provided a "first-order" correction of the initial retention mismatch observed for the nominally identical columns, while additional alignment via COW was required to optimize sample classification by PCA.

Simulating comprehensive two-dimensional gas chromatography mass spectrometry data with realistic run-to-run shifting to evaluate the robustness of tile-based Fisher ratio analysis

Untargeted analysis of comprehensive two-dimensional (2D) gas chromatography time-of-flight mass spectrometry (GC×GC-TOFMS) data has the potential to be hindered by run-to-run retention time shifting. To address this challenge, tile-based Fisher ratio (F-ratio) analysis (FRA) has been developed, which utilizes a supervised, untargeted approach involving a chromatographic segmentation routine termed "tiling" combined with the ANOVA F-ratio statistic to discover class-distinguishing analytes while minimizing false positives arising from shifting. The tiling algorithm is designed to account for retention shifting in both separation dimensions. Although applications of FRA have been reported, there remains a need to thoroughly evaluate the robustness of FRA for different levels of run-to-run retention shifting in order to broaden the scope of its application. To this end, a novel method of simulating GC×GC-TOFMS chromatograms with realistic run-to-run shifting is presented by random generation of low-frequency "shift functions". The dimensionless retention-time precision, <δ_r>, which is four times the standard deviation in retention time normalized to the peak width-at-base is used as a key modeling variable along with the 2D chromatographic saturation, α_e,2D, and within-class relative standard deviation in peak area, RSD_wc. We demonstrate that all three of these variables operate together to impact true positive discovery. To quantify the "success" of true positive discovery, GC×GC-TOFMS datasets for various combinations of <δ_r>, α_e,2D, and RSD_wc were simulated and then analyzed by FRA using a wide range of relative tile areas (RTA), which is a dimensionless measure of tile size. Since each hit in the FRA hit list was known a priori as either a true or false positive based on the simulation inputs, receiver operating characteristic (ROC) curves were readily constructed. Then, the area under the ROC curve (AUROC) was used as a metric for discovery "success" for various combinations of the modeling variables. Based on the results of this study, recommendations for tile size selection and experimental design are provided, and further supported by comparison to previous tile-based FRA applications. For instance, values for <δ_r>, α_e,2D, and RSD_wc obtained from a GC×GC-TOFMS dataset of yeast metabolites suggested an optimum RTA of 6.25, corresponding closely to the RTA of 4.00 employed in the study, implying the simulation results obtained here can be generalized to real datasets.

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.

Development of variance rank initiated-unsupervised sample indexing for gas chromatography-mass spectrometry analysis

Traditional non-targeted chemometric workflows for gas chromatography-mass spectrometry (GC-MS) data rely on using supervised methods, which requires a priori knowledge of sample class membership. Herein, we propose a simple, unsupervised chemometric workflow known as variance rank initiated-unsupervised sample indexing (VRI-USI). VRI-USI discovers analyte peaks exhibiting high relative variance across all samples, followed by k-means clustering on the individual peaks. Based upon how the samples cluster for a given peak, a sample index assignment is provided. Using a probabilistic argument, if the same sample index assignment appears for several discovered peaks, then this outcome strongly suggests that the samples are properly classified by that particular sample index assignment. Thus, relevant chemical differences between the samples have been discovered in an unsupervised fashion. The VRI-USI workflow is demonstrated on three, increasingly difficult datasets: simulations, yeast metabolomics, and human cancer metabolomics. For simulated GC-MS datasets, VRI-USI discovered 85-90% of analytes modeled to vary between sample classes. Nineteen out of 53 peaks in the peak table developed for the yeast metabolome dataset had the same sample index assignments, indicating that those indices are most likely due to class-distinguishing chemical differences. A t-test revealed that 22 out of 53 peaks were statistically significant (p < 0.05) when using those sample index assignments. Likewise, for the human cancer metabolomics study, VRI-USI discovered 25 analytes that were statistically different (p < 0.05) using the sample index assignments determined to highlight meaningful sample-based differences. For all datasets, the sample index assignments that were deduced from VRI-USI were the correct class-based difference when using prior knowledge. VRI-USI holds promise as an exploratory data analysis workflow for studies in which analysts do not readily have a priori class information or want to uncover the underlying nature of their dataset.

The Often-Overlooked Power of Summary Statistics in Exploratory Data Analysis: Comparison of Pattern Recognition Entropy (PRE) to Other Summary Statistics and Introduction of Divided Spectrum-PRE (DS-PRE)

Unsupervised exploratory data analysis (EDA) is often the first step in understanding complex data sets. While summary statistics are among the most efficient and convenient tools for exploring and describing sets of data, they are often overlooked in EDA. In this paper, we show multiple case studies that compare the performance, including clustering, of a series of summary statistics in EDA. The summary statistics considered here are pattern recognition entropy (PRE), the mean, standard deviation (STD), 1-norm, range, sum of squares (SSQ), and X⁴, which are compared with principal component analysis (PCA), multivariate curve resolution (MCR), and/or cluster analysis. PRE and the other summary statistics are direct methods for analyzing data-they are not factor-based approaches. To quantify the performance of summary statistics, we use the concept of the "critical pair," which is employed in chromatography. The data analyzed here come from different analytical methods. Hyperspectral images, including one of a biological material, are also analyzed. In general, PRE outperforms the other summary statistics, especially in image analysis, although a suite of summary statistics is useful in exploring complex data sets. While PRE results were generally comparable to those from PCA and MCR, PRE is easier to apply. For example, there is no need to determine the number of factors that describe a data set. Finally, we introduce the concept of divided spectrum-PRE (DS-PRE) as a new EDA method. DS-PRE increases the discrimination power of PRE. We also show that DS-PRE can be used to provide the inputs for the k-nearest neighbor (kNN) algorithm. We recommend PRE and DS-PRE as rapid new tools for unsupervised EDA.

Sediment quality assessment combining chemical and biological (non)target analysis

Aquatic sediments act as a storage for diverse mixtures of organic and inorganic contaminants. Nevertheless, most evaluations of contaminated sediments have been limited to the assessment of concentrations of target compounds and lethal effects on some test species. To identify the organic contaminants causing sub-lethal effects of contaminated sediment, this study combined chemical and biological (non)target analysis involving comprehensive two-dimensional gas chromatography coupled with a time-of-flight Mass Spectrometer (GCxGC/ToF-MS) analysis, embryonic malformation and high-throughput sequencing (RNA-seq) analysis on developing flounder. Polycyclic aromatic hydrocarbons were more abundant in the sediment extract of Yeongil Bay (SEY), while Jinhae Bay (SEJ) was contaminated with a large amount of unidentified chemicals. The unidentified chemicals of SEJ included branched alkanes, oxygenated cycloalkanes, heterogeneous hydrocarbons, and other unknown compounds. Percentage of pericardial edema was the highest in embryonic flounder exposed to SEY. Consistent with the morphogenesis results, the expression level of genes related to heart formation including the nkx2.5 and robo1 was greater in embryonic flounder exposed to SEY. In the analyses of differential gene expression profiles (cutoff P < 0.05), by RNA-seq, embryos exposed to SEJ showed changes related to cell differentiation, cell part morphogenesis, neurogenesis, and neuron development. Genes related to neurogenesis and positive regulation of molecular functions variated significantly in embryos exposed to SEY. These results demonstrated the advantages of combining target and non-target analysis to accurately evaluate the major chemical groups causing sediment toxicity. Therefore, this work provided a useful approach to tracking and revealing the causes of toxic effects and identifying potential toxic mechanisms.

Determination of the Signal-To-Noise Ratio Enhancement in Comprehensive Three-Dimensional Gas Chromatography

We investigate the extent to which comprehensive three-dimensional gas chromatography (GC³) provides a signal enhancement (SE) and a signal-to-noise ratio enhancement (S/N_Rel) relative to one-dimensional (1D)-GC. Specifically, the SE is defined as the ratio of the tallest ³D peak height from the GC³ separation to the 1D peak height from the unmodulated 1D-GC separation. A model is proposed which allows the analyst to predict the theoretically attainable SE (SE_T) based upon the peak width and sampling density inputs. The model is validated via comparison of the SE_T to the experimentally measured SE (SE_M) obtained using total-transfer GC³ (100% duty cycle for both modulators) with time-of-flight mass spectrometry detection. Two experimental conditions were studied using the same GC³ column set, differing principally in the modulation period from the ¹D to ²D columns: 4 s versus 8 s. Under the first set of conditions, the average SE_M was 97 (±22), in excellent agreement with the SE_T of 97 (±18). The second set of conditions improved the average SE_M to 181 (±27), also in agreement with the average SE_T of 176 (±26). The average S/N_Rel following correction for the mass spectrum acquisition frequency was 38.8 (±11.2) and 59.0 (±27.2) for the two sets of conditions. The enhancement in S/N is largely attributed to moving the signal to a higher frequency domain where the impact of "low frequency" noise is less detrimental. The findings here provide strong evidence that GC³ separations can provide enhanced detectability relative to 1D-GC and comprehensive two-dimensional gas chromatography (GC×GC) separations.

Prediction by statistical overlap theory of fraction of baseline occupied by chromatographic peaks

The average minimum resolution required for separating adjacent single-component peaks (SCPs) in one-dimensional chromatograms is an important metric in statistical overlap theory (SOT). However, its value changes with changing chromatographic conditions in non-intuitive ways, when SOT predicts the average number of peaks (maxima). A more stable and easily understood value of resolution is obtained on making a different prediction. A general equation is derived for the sum of all separated and superposed widths of SCPs in a chromatogram. The equation is a function of the saturation α, a metric of chromatographic crowdedness, and is expressed in dimensionless form by dividing by the duration of the chromatogram. This dimensionless function, f(α), is also the cumulative distribution function of the probability of separating adjacent SCPs. Simulations based on the clustering of line segments representing SCPs verify expressions for f(α) calculated from five functions for the distribution of intervals between adjacent SCPs. Synthetic chromatograms are computed with different saturations, distributions of intervals, and distribution of SCP amplitudes. The chromatograms are analyzed by calculating the sum of the widths of peaks at different relative responses, dividing the sum by the duration of the chromatograms, and graphing the reduced sum against relative response. For small values of relative response, the reduced sum approaches the fraction of baseline that is occupied by chromatographic peaks. This fraction can be identified with f(α), if the saturation α is defined with the average minimum resolution equaling 1.5. The identification is general and is independent of the saturation, the interval distribution, or the amplitude distribution. This constant value of resolution corresponds to baseline resolution, which simplifies the interpretation of SOT.