Trusting our machines: validating machine learning models for single-molecule transport experiments.

In this tutorial review, we will describe crucial aspects related to the application of machine learning to help users avoid the most common pitfalls. The examples we present will be based on data from the field of molecular electronics, specifically single-molecule electron transport experiments, but the concepts and problems we explore will be sufficiently general for application in other fields with similar data. In the first part of the tutorial review, we will introduce the field of single-molecule transport, and provide an overview of the most common machine learning algorithms employed. In the second part of the tutorial review, we will show, through examples grounded in single-molecule transport, that the promises of machine learning can only be fulfilled by careful application. We will end the tutorial review with a discussion of where we, as a field, could go from here.

A Metabolomic Approach to Beer Characterization.

The consumers' interest towards beer consumption has been on the rise during the past decade: new approaches and ingredients get tested, expanding the traditional recipe for brewing beer. As a consequence, the field of "beeromics" has also been constantly growing, as well as the demand for quick and exhaustive analytical methods. In this study, we propose a combination of nuclear magnetic resonance (NMR) spectroscopy and chemometrics to characterize beer. 1H-NMR spectra were collected and then analyzed using chemometric tools. An interval-based approach was applied to extract chemical features from the spectra to build a dataset of resolved relative concentrations. One aim of this work was to compare the results obtained using the full spectrum and the resolved approach: with a reasonable amount of time needed to obtain the resolved dataset, we show that the resolved information is comparable with the full spectrum information, but interpretability is greatly improved.

Untargeted Metabolomic Profile for the Detection of Prostate Carcinoma-Preliminary Results from PARAFAC2 and PLS-DA Models.

Prostate-specific antigen (PSA) is the main biomarker for the screening of prostate cancer (PCa), which has a high sensibility (higher than 80%) that is negatively offset by its poor specificity (only 30%, with the European cut-off of 4 ng/mL). This generates a large number of useless biopsies, involving both risks for the patients and costs for the national healthcare systems. Consequently, efforts were recently made to discover new biomarkers useful for PCa screening, including our proposal of interpreting a multi-parametric urinary steroidal profile with multivariate statistics. This approach has been expanded to investigate new alleged biomarkers by the application of untargeted urinary metabolomics. Urine samples from 91 patients (43 affected by PCa; 48 by benign hyperplasia) were deconjugated, extracted in both basic and acidic conditions, derivatized with different reagents, and analyzed with different gas chromatographic columns. Three-dimensional data were obtained from full-scan electron impact mass spectra. The PARADISe software, coupled with NIST libraries, was employed for the computation of PARAFAC2 models, the extraction of the significative components (alleged biomarkers), and the generation of a semiquantitative dataset. After variables selection, a partial least squares-discriminant analysis classification model was built, yielding promising performances. The selected biomarkers need further validation, possibly involving, yet again, a targeted approach.

Forecasting Chronic Diseases Using Data Fusion.

Data fusion, that is, extracting information through the fusion of complementary data sets, is a topic of great interest in metabolomics because analytical platforms such as liquid chromatography-mass spectrometry (LC-MS) and nuclear magnetic resonance (NMR) spectroscopy commonly used for chemical profiling of biofluids provide complementary information. In this study, with a goal of forecasting acute coronary syndrome (ACS), breast cancer, and colon cancer, we jointly analyzed LC-MS, NMR measurements of plasma samples, and the metadata corresponding to the lifestyle of participants. We used supervised data fusion based on multiple kernel learning and exploited the linearity of the models to identify significant metabolites/features for the separation of healthy referents and the cases developing a disease. We demonstrated that (i) fusing LC-MS, NMR, and metadata provided better separation of ACS cases and referents compared with individual data sets, (ii) NMR data performed the best in terms of forecasting breast cancer, while fusion degraded the performance, and (iii) neither the individual data sets nor their fusion performed well for colon cancer. Furthermore, we showed the strengths and limitations of the fusion models by discussing their performance in terms of capturing known biomarkers for smoking and coffee. While fusion may improve performance in terms of separating certain conditions by jointly analyzing metabolomics and metadata sets, it is not necessarily always the best approach as in the case of breast cancer.

Using PAT to accelerate the transition to continuous API manufacturing.

Significant improvements can be realized by converting conventional batch processes into continuous ones. The main drivers include reduction of cost and waste, increased safety, and simpler scale-up and tech transfer activities. Re-designing the process layout offers the opportunity to incorporate a set of process analytical technologies (PAT) embraced in the Quality-by-Design (QbD) framework. These tools are used for process state estimation, providing enhanced understanding of the underlying variability in the process impacting quality and yield. This work describes a road map for identifying the best technology to speed-up the development of continuous processes while providing the basis for developing analytical methods for monitoring and controlling the continuous full-scale reaction. The suitability of in-line Raman, FT-infrared (FT-IR), and near-infrared (NIR) spectroscopy for real-time process monitoring was investigated in the production of 1-bromo-2-iodobenzene. The synthesis consists of three consecutive reaction steps including the formation of an unstable diazonium salt intermediate, which is critical to secure high yield and avoid formation of by-products. All spectroscopic methods were able to capture critical information related to the accumulation of the intermediate with very similar accuracy. NIR spectroscopy proved to be satisfactory in terms of performance, ease of installation, full-scale transferability, and stability to very adverse process conditions. As such, in-line NIR was selected to monitor the continuous full-scale production. The quantitative method was developed against theoretical concentration values of the intermediate since representative sampling for off-line reference analysis cannot be achieved. The rapid and reliable analytical system allowed the following: speeding up the design of the continuous process and a better understanding of the manufacturing requirements to ensure optimal yield and avoid unreacted raw materials and by-products in the continuous reactor effluent. Graphical Abstract Using PAT to accelerate the transition to continuous API manufacturing.

The effects of water and dairy drinks on dietary patterns in overweight adolescents.

The aim was to investigate the effects of increased water or dairy intake on total intake of energy, nutrients, foods and dietary patterns in overweight adolescents in the Milk Components and Metabolic Syndrome (MoMS) study (n=173). Participants were randomly assigned to consume 1l/d of skim milk, whey, casein or water for 12 weeks. A decrease in the dietary pattern called Convenience Food, identified by principal component analysis, was observed during the intervention both in the water and dairy groups. Total energy intake decreased by 990.9 kJ/d (236.8 kcal/d) in the water group but was unchanged in the dairy group during intervention. To conclude, an extra intake of fluid seems to favourably affect the rest of the diet by decreasing the intake of convenience foods, including sugar-sweetened beverages. A low energy drink, such as water, seems advantageous considering the total energy intake in these overweight adolescents. This study is registered at clinicaltrials.gov (NCT00785499).

Direct, simultaneous quantification of fructooligosaccharides by FT-MIR ATR spectroscopy and chemometrics for rapid identification of superior, engineered ß-fructofuranosidases.

Fructooligosaccharides (FOS) are popular components of functional foods produced by the enzymatic transfer of fructose units to sucrose. Improving ß-fructofuranosidase traits by protein engineering is restricted by the absence of a rapid, direct screening method for the fructooligosaccharide products produced by enzyme variants. The use of standard high-performance liquid chromatography (HPLC) methods involves time-consuming sample preparation and chromatographic and data analysis steps. To overcome these limitations, this work presents a rapid method for screening ß-fructofuranosidase variant libraries using Fourier transform mid-infrared attenuated total reflectance (FT-MIR ATR) spectroscopy and calibration using partial least squares (PLS) regression. The method offers notable improvements in terms of sample analysis times and cost, with the added benefit of the absence of toxic eluents. Wavenumber interval selection methods were tested to develop optimised PLS regression models that were successfully applied to quantify of glucose, fructose, sucrose, 1-kestose and nystose, the substrates and products of ß-fructofuranosidase activity. To the best of our knowledge, this is the first report on the use of infrared spectroscopy and PLS calibration for the quantification of 1-kestose and nystose. Independent test set-validated results indicated that optimal wavenumber selection by interval PLS (iPLS) served to provide the best models for all sugars, bar glucose. Application of this screening method will facilitate the engineering of ß-fructofuranosidases and other glycosyltransferase enzymes by random mutagenesis strategies, as it provides, for the first time, a rapid, direct assay for transferase products that may be adapted to a high-throughput set-up.

Structure-revealing data fusion.

BACKGROUND: Analysis of data from multiple sources has the potential to enhance knowledge discovery by capturing underlying structures, which are, otherwise, difficult to extract. Fusing data from multiple sources has already proved useful in many applications in social network analysis, signal processing and bioinformatics. However, data fusion is challenging since data from multiple sources are often (i) heterogeneous (i.e., in the form of higher-order tensors and matrices), (ii) incomplete, and (iii) have both shared and unshared components. In order to address these challenges, in this paper, we introduce a novel unsupervised data fusion model based on joint factorization of matrices and higher-order tensors. RESULTS: While the traditional formulation of coupled matrix and tensor factorizations modeling only shared factors fails to capture the underlying structures in the presence of both shared and unshared factors, the proposed data fusion model has the potential to automatically reveal shared and unshared components through modeling constraints. Using numerical experiments, we demonstrate the effectiveness of the proposed approach in terms of identifying shared and unshared components. Furthermore, we measure a set of mixtures with known chemical composition using both LC-MS (Liquid Chromatography - Mass Spectrometry) and NMR (Nuclear Magnetic Resonance) and demonstrate that the structure-revealing data fusion model can (i) successfully capture the chemicals in the mixtures and extract the relative concentrations of the chemicals accurately, (ii) provide promising results in terms of identifying shared and unshared chemicals, and (iii) reveal the relevant patterns in LC-MS by coupling with the diffusion NMR data. CONCLUSIONS: We have proposed a structure-revealing data fusion model that can jointly analyze heterogeneous, incomplete data sets with shared and unshared components and demonstrated its promising performance as well as potential limitations on both simulated and real data.

An automated method for baseline correction, peak finding and peak grouping in chromatographic data.

An automated method (FastChrom) for baseline correction, peak detection and assignment (grouping) of similar peaks across samples has been developed. The method has been tested both on artificial data and a dataset obtained from gas chromatograph analysis of wine samples. As part of the automated approach, a new method for baseline estimation has been developed and compared with other methods. FastChrom has been shown to perform at least as well as conventional software. However, compared to other approaches, FastChrom finds more peaks in the chromatograms and not only those with retention times defined by the user. FastChrom is fast and easy to use and offers the possibility of applying a retention time index which facilitates the identification of peaks and the comparison between experiments.

PARASIAS: A new method for analyzing higher-order tensors with shifting profiles.

Higher-order tensor data analysis has been extensively employed to understand complicated data, such as multi-way GC-MS data in untargeted/targeted analysis. However, the analysis can be complicated when one of the modes shifts e.g., the elution profiles of specific compounds often with respect to retention time; something which violates the assumptions of more traditional models. In this paper, we introduce a new analysis method named PARASIAS for analyzing shifted higher-order tensor data by combining spectral transformation and the simple PARAFAC modeling. The proposed method is validated by applications on both simulated and real multi-way datasets. Compared to the state-of-art PARAFAC2 model, the results indicate that fitting of PARASIAS is 13 times faster on simulated datasets and more than eight times faster on average on the real datasets studied. PARASIAS has significant advantages in terms of model simplicity, convergence speed, the robustness to shift changes in the data, the ability to impose non-negativity constraint on the shift mode and the possibility of easily extending to data with multiple shift modes. However, the resolved profiles of PARASIAS model are always a little worse when the number of components in the data are larger than three and without using additional factors in PARASIAS model. In such cases, more components are necessary for PARASIAS to model the data than that would be needed e.g., by PARAFAC2. The reason for this is also discussed in this work.

Excitation-Emission Matrix Fluorescence Spectroscopy Coupled with PARAFAC Modeling for Viability Prediction of Cells.

Cell-based sensors and assays have great potential in bioanalysis, drug discovery screening, and biochemical mechanisms research. The cell viability tests should be fast, safe, reliable, and time- and cost-effective. Although methods stated as "gold standards", such as MTT, XTT, and LDH assays, usually fulfill these assumptions, they also show some limitations. They can be time-consuming, labor-intensive, and prone to errors and interference. Moreover, they do not enable the observation of the cell viability changes in real-time, continuously, and nondestructively. Therefore, we propose an alternative method of viability testing: native excitation-emission matrix fluorescence spectroscopy coupled with parallel factor analysis (PARAFAC), which is especially advantageous for cell monitoring due to its noninvasiveness and nondestructiveness and because there is no need for labeling and sample preparation. We demonstrate that our approach provides accurate results with even better sensitivity than the standard MTT test. With PARAFAC, it is possible to study the mechanism of the observed cell viability changes, which can be directly linked to increasing/decreasing fluorophores in the cell culture medium. The resulting parameters of the PARAFAC model are also helpful in establishing a reliable regression model for accurate and precise determination of the viability in A375 and HaCaT-adherent cell cultures treated with oxaliplatin.

Classification of bloodstains deposited at different times on floor tiles using hierarchical modelling and a handheld NIR spectrometer.

Bloodstains are commonly encountered at crime scenes, especially on floor tiles, and can be deposited over different periods and intervals. Therefore, it is crucial to develop techniques that can accurately identify bloodstains deposited at different times. This study builds upon a previous investigation and aims to enhance the performance of three distinct hierarchical models (HMs) designed to differentiate and identify stains of human blood (HB), animal blood (AB), and common false positives (CFPs) on nine different types of floor tiles. Soft Independent Modeling Class Analogies (SIMCA), and Partial Least Squares-Discriminant Analysis (PLS-DA) were employed as decision rules in this process. The originally published model was constructed using a training set that included samples with a known time of deposit of six days. This model was then tested to predict samples with various deposition times, including human blood samples aged for 0, 1, 9, 20, 30, and 162 days, as well as animal blood samples aged for 0, 1, 10, 13, 20, 29, 105, and 176 days. To improve the identification of human blood, the models were modified by adding zero-day and one-day-old bloodstains to the original training set. All models showed improvement when fresher samples were included in the training set. The best results were achieved with the hierarchical model that used partial least squares-discriminant analysis as the second decision rule and incorporated one-day-old samples in the training set. This model yielded sensitivity values above 0.92 and specificity values above 0.7 for samples aged between zero and 30 days.

Colorimetric pH indicators based on well-defined amylose and amylopectin matrices enriched with anthocyanins from red cabbage.

This study aimed to prepare a novel colorimetric indicator film from virtually pure (99 %) amylose (AM) and anthocyanins extracted from red cabbage (RCA). The AM used was a unique engineered bulk material extracted from transgenic barley grains. Films produced by solution casting were compared to normal barely starch (NB) and pure barley amylopectin (AP), with amylose contents of 30 % and 0 %, respectively. The pH-indicator films were produced by incorporation of RCA into the different starch support matrices with different amylose contents. Barrier, thermal, and mechanical properties, photo degradation stability, and release behavior data revealed that RCA interact differently through the glucan matrices. Microstructural observations showed that RCA were evenly dispersed in the glucan matrix, and AM+RCA indicator films showed high UV-barrier and mechanical performance over normal starch. FTIR revealed that RCA was properly affected by the AM matrix. Moreover, the AM+RCA films showed sensitive color changes in the pH range (2-11) and a predominant Fickian diffusion release mechanism for RCA. This study provides for the first time data regarding AM films with RCA and their promising potential for application as support matrices in responsive food and other industrial biodegradable packaging materials.

Using fluorescence excitation-emission matrices to predict bitterness and pungency of virgin olive oil: A feasibility study.

Unlike other food products, virgin olive oil must undergo an organoleptic assessment that is currently based on a trained human panel, which presents drawbacks that might affect the efficiency and robustness. Therefore, disposing of instrumental methods that could serve as screening tools to support sensory panels is of paramount importance. The present work aimed to explore excitation-emission fluorescence spectroscopy (EEFS) to predict bitterness and pungency, since both attributes are related with fluorophore compounds, such as polar phenols. Bitterness and pungency intensities of 250 samples were provided by an official sensory panel and used to build and compare partial least squares regressions (PLSR) with the excitation-emission matrix. Both PARAFAC scores and two-way unfolded data led to successful PLSR. The most relevant PARAFAC scores agreed with virgin olive oil phenolic spectra, evidencing that EEFS would be the fit-for-purpose screening tool to support the sensory panel.

Hierarchical classification models and Handheld NIR spectrometer to human blood stains identification on different floor tiles.

One of the most important types of evidence in certain criminal investigations is traces of human blood. For a detailed investigation, blood samples must be identified and collected at the crime scene. The present study aimed to evaluate the potential of the identification of human blood in stains deposited on different types of floor tiles (five types of ceramics and four types of porcelain tiles) using a portable NIR instrument. Hierarchical models were developed by combining multivariate analysis techniques capable of identifying traces of human blood (HB), animal blood (AB) and common false positives (CFP). The spectra of the dried stains were obtained using a portable MicroNIR spectrometer (Viavi). The hierarchical models used two decision rules, the first to separate CFP and the second to discriminate HB from AB. The first decision rule, used to separate the CFP, was based on the Q-Residual criterion considering a PCA model. For the second rule, used to discriminate HB and AB, the Q-Residual criterion were tested as obtained from a PCA model, a One-Class SIMCA model, and a PLS-DA model. The best results of sensitivity and specificity, both equal to 100%, were obtained when a PLS-DA model was employed as the second decision rule. The hierarchical classification models built for these same training sets using a PCA or SIMCA model also obtained excellent sensitivity results for HB classification, with values above 94% and 78% of specificity. No CFP samples were misclassified. Hierarchical models represent a significant advance as a methodology for the identification of human blood stains at crime scenes.

Automatic and non-targeted analysis of the volatile profile of natural and alkalized cocoa powders using SBSE-GC-MS and chemometrics.

A total of 56 key volatile compounds present in natural and alkalized cocoa powders have been rapidly evaluated using a non-target approach using stir bar sorptive extraction gas chromatography mass spectrometry (SBSE-GC-MS) coupled to Parallel Factor Analysis 2 (PARAFAC2) automated in PARADISe. Principal component analysis (PCA) explained 80% of the variability of the concentration, in four PCs, which revealed specific groups of volatile characteristics. Partial least squares discriminant analysis (PLS-DA) helped to identify volatile compounds that were correlated to the different degrees of alkalization. Dynamics between compounds such as the acetophenone increasing and toluene and furfural decreasing in medium and strongly alkalized cocoas allowed its differentiation from natural cocoa samples. Thus, the proposed comprehensive analysis is a useful tool for understanding volatiles, e.g., for the quality control of cocoa powders with significant time and costs savings.

Spider web biomonitoring: A cost-effective source apportionment approach for urban particulate matter.

Elevated levels of particulate matter (PM) in urban atmospheres are one of the major environmental challenges of the Anthropocene. To effectively lower those levels, identification and quantification of sources of PM is required. Biomonitoring methods are helpful tools to tackle this problem but have not been fully established yet. An example is the sampling and subsequent analysis of spider webs to whose adhesive surface dust particles can attach. For a methodical inspection, webs of orb-weaving spiders were sampled repeatedly from 2016 to 2018 at 22 locations in the city of Jena, Germany. Contents of Ag, Al, As, B, Ba, Ca, Cd, Co, Cr, Cs, Cu, Fe, K, La, Li, Mg, Mn, Mo, Na, Ni, P, Pb, Rb, S, Sb, Si, Sn, Sr, Th, Ti, V, Y, Zn and Zr were determined in the samples using inductively coupled plasma-mass spectrometry (ICP-MS) and inductively coupled plasma-optical emission spectroscopy (ICP-OES) after aqua regia digestion. Multivariate statistical methods were applied for a detailed evaluation. A combination of cluster analysis and principal component analysis allows for the clear identification of three main sources in the study area: brake wear from car traffic, abrasion of tram/train tracks and particles of geogenic origin. Quantitative source contributions reveal that high amounts of most of the metals are derived from a combination of brake wear and geogenic particles, the latter of which are likely resuspended by moving vehicles. This emphasizes the importance of non-exhaust particles connected to road traffic. Once a source identification has been performed for an area of interest, classification models can be applied to assess air quality for further samples from within the whole study area, offering a tool for air quality assessment. The general validity of this approach is demonstrated using samples from other locations.

Multivariate evaluation of pharmacological responses in early clinical trials - a study of rIL-21 in the treatment of patients with metastatic melanoma.

AIMS: Evaluation of the utility of multivariate data analysis in early clinical drug development. METHODS: A multivariate chemometric approach was developed and applied for evaluating clinical laboratory parameters and biomarkers obtained from two clinical trials investigating recombinant human interleukin-21 (rIL-21) in the treatment of patients with malignant melanoma. The Phase I trial was an open-label, first-human dose escalation safety and tolerability trial with two separate dosing regimens; six cycles of thrice weekly (3/w) vs. three cycles of daily dosing for 5 days followed by 9 days of rest (5+9) in a total of 29 patients. The Phase II trial investigated efficacy and safety of the '5+9' regimen in 24 patients. RESULTS: From the Phase I trial, separate pharmacological patterns were observed for each regimen, clearly reflecting distinct properties of the two regimens. Relations between individual laboratory parameters were visualized and shown to be responsive to rIL-21 dosing. In particular, novel systematic pharmacological effects on liver function parameters as well as a bell-shaped dose-response relationship of the overall pharmacological effects were depicted. In validation of the method, multivariate pharmacological patterns discovered in the Phase I trial could be reproduced by the dataset from the Phase II trial, but not from univariate exploration of the Phase I trial. CONCLUSIONS: The new data analytical approach visualized novel correlations between laboratory parameters that points to specific pharmacological properties. This multivariate chemometric data analysis offers a novel robust, comprehensive and intuitive tool to reveal early pharmacological responses and guide selection of dose regimens.

Can We Trust Score Plots?

In this paper, we discuss the validity of using score plots of component models such as partial least squares regression, especially when these models are used for building classification models, and models derived from partial least squares regression for discriminant analysis (PLS-DA). Using examples and simulations, it is shown that the currently accepted practice of showing score plots from calibration models may give misleading interpretations. It is suggested and shown that the problem can be solved by replacing the currently used calibrated score plots with cross-validated score plots.

Feasibility of serodiagnosis of ovarian cancer by mass spectrometry.

The emergence of new biological disease markers from mass spectrometric studies of serum proteomes has been quite limited. There are challenges regarding the analytical and statistical procedures, preanalytical variability, and study designs. In this serological study of ovarian cancer, we apply classification methods in a strictly designed study with standardized sample collection procedures. A total of 265 sera from women admitted with symptoms of a pelvic mass were used for model building. We developed a rigorous approach for building classification models suitable for the highly multivariate data and illustrate how to evaluate and ensure data quality and optimize data preprocessing and data reduction. We document time dependent changes in peak profiles up to 15 months after sampling even when storing samples at -20 degrees C. The developed classification model was validated using completely independent samples, and a cross validation procedure which we call cross model validation was applied to get realistic performance values. The best models were able to classify with 79% specificity and 56% sensitivity, i.e., an analytical accuracy of 68%. However, the existing serum marker (CA-125) alone gave a better analytical accuracy (81%) in the same sample set. Also, the combination of mass spectrometric data and levels of CA-125 data did not improve the predictive performance of models. In conclusion, proteomic approaches to biomarker discovery are not necessarily yielding straightforward diagnostic leads but lay the foundation for more work.