Fast and automated biomarker detection in breath samples with machine learning.

Volatile organic compounds (VOCs) in human breath can reveal a large spectrum of health conditions and can be used for fast, accurate and non-invasive diagnostics. Gas chromatography-mass spectrometry (GC-MS) is used to measure VOCs, but its application is limited by expert-driven data analysis that is time-consuming, subjective and may introduce errors. We propose a machine learning-based system to perform GC-MS data analysis that exploits deep learning pattern recognition ability to learn and automatically detect VOCs directly from raw data, thus bypassing expert-led processing. We evaluate this new approach on clinical samples and with four types of convolutional neural networks (CNNs): VGG16, VGG-like, densely connected and residual CNNs. The proposed machine learning methods showed to outperform the expert-led analysis by detecting a significantly higher number of VOCs in just a fraction of time while maintaining high specificity. These results suggest that the proposed novel approach can help the large-scale deployment of breath-based diagnosis by reducing time and cost, and increasing accuracy and consistency.

Volatile atmospheric pressure chemical ionisation mass spectrometry headspace analysis of E. coli and S. aureus.

Identifying the characteristics of bacterial species can improve treatment outcomes and mass spectrometry methods have been shown to be capable of identifying biomarkers of bacterial species. This study is the first to use volatile atmospheric pressure chemical ionisation mass spectrometry to directly and non-invasively analyse the headspace of E. coli and S. aureus bacterial cultures, enabling major biological classification at species level (Gram negative/positive respectively). Four different protocols were used to collect data, three utilising discrete 5 min samples taken between 2 and 96 h after inoculation and one method employing 24 h continuous sampling. Characteristic marker ions were found for both E. coli and S. aureus. A model to distinguish between sample types was able to correctly identify the bacteria samples after sufficient growth (24-48 h), with similar results obtained across different sampling methods. This demonstrates that this is a robust method to analyse and classify bacterial cultures accurately and within a relevant time frame, offering a promising technique for both clinical and research applications.

Associations between maternal characteristics and pharmaceutical treatment of gestational diabetes: an analysis of the UK Born in Bradford (BiB) cohort study.

OBJECTIVES: To identify the maternal characteristics associated with pharmaceutical treatment of gestational diabetes mellitus (GDM). DESIGN: Prospective birth cohort study. SETTING: Bradford, UK. PARTICIPANTS: 762 women from the Born in Bradford (BiB) cohort who were treated for GDM in a singleton pregnancy. BiB cohort participants were recruited from 2007 to 2010. All women booked for delivery were screened for GDM between 26 and 28 weeks of gestation using a 75 g 2-hour oral glucose tolerance test (OGTT). OUTCOME MEASURE: GDM treatment type: lifestyle changes advice (lifestyle changes), lifestyle changes advice with supplementary insulin (insulin) and lifestyle changes advice with supplementary metformin (metformin). RESULTS: 244 (32%) women were prescribed lifestyle changes advice alone while 518 (68%) were offered supplemental pharmaceutical treatment. The odds of receiving pharmaceutical treatment relative to lifestyle changes advice alone were increased for mothers who were obese (OR 4.6, 95% CI 2.8 to 7.5), those who smoked (OR 2.6, 95% CI 1.2 to 5.5) and had higher fasting glucose levels at OGTT (OR 2.1, 95% CI 1.6 to 2.7). The odds of being prescribed pharmaceutical treatment rather than lifestyle changes advice were lower for Pakistani women (OR 0.7, 95% CI 0.4 to 1.0)) than White British women. Relative to insulin treatment, metformin was more likely to be offered to obese women than normal weight women (relative risk ratio, RRR 3.2, 95% CI 1.3 to 7.8) and less likely to be prescribed to women with higher fasting glucose concentrations at OGTT (RRR 0.3, 95% CI 0.2 to 0.6). CONCLUSIONS: In the BiB cohort, GDM pharmaceutical treatment tended to be prescribed to women who were obese, White British, who smoked and had more severe hyperglycaemia. The characteristics of metformin-treated mothers differed from those of insulin-treated mothers as they were more likely to be obese but had lower glucose concentrations at diagnosis.

Automated Searching and Identification of Self-Organized Nanostructures.

Currently, researchers spend significant time manually searching through large volumes of data produced during scanning probe imaging to identify specific patterns and motifs formed via self-assembly and self-organization. Here, we use a combination of Monte Carlo simulations, general statistics, and machine learning to automatically distinguish several spatially correlated patterns in a mixed, highly varied data set of real AFM images of self-organized nanoparticles. We do this regardless of feature-scale and without the need for manually labeled training data. Provided that the structures of interest can be simulated, the strategy and protocols we describe can be easily adapted to other self-organized systems and data sets.

A Tunable Three-Dimensional Printed Microfluidic Resistive Pulse Sensor for the Characterization of Algae and Microplastics.

Technologies that can detect and characterize particulates in liquids have applications in health, food, and environmental monitoring. Simply counting the numbers of cells or particles is not sufficient for most applications; other physical properties must also be measured. Typically, it is necessary to compromise between the speed of a sensor and its chemical and biological specificity. Here, we present a low-cost and high-throughput multiuse counter that classifies a particle's size, concentration, and shape. We also report how the porosity/conductivity or the particle can influence the signal. Using an additive manufacturing process, we have assembled a reusable flow resistive pulse sensor capable of being tuned in real time to measure particles from 2 to 30 µm across a range of salt concentrations, i.e., 2.5 × 10-4 to 0.1 M. The device remains stable for several days with repeat measurements. We demonstrate its use for characterizing algae with spherical and rod structures as well as microplastics shed from tea bags. We present a methodology that results in a specific signal for microplastics, namely, a conductive pulse, in contrast to particles with smooth surfaces such as calibration particles or algae, allowing the presence of microplastics to be easily confirmed and quantified. In addition, the shapes of the signal and of the particle are correlated, giving an extra physical property to characterize suspended particulates. The technology can rapidly screen volumes of liquid, 1 mL/min, for the presence of microplastics and algae.

Breath selection methods for compact mass spectrometry breath analysis.

Compact mass spectrometry (CMS) is a versatile and transportable analytical instrument that has the potential to be used in clinical settings to quickly and non-invasively detect a wide range of relevant conditions from breath samples. The purpose of this study is to optimise data preprocessing protocols by three proposed methods of breath sampling, using the CMS. It also lays out a general framework for which data processing methods can be evaluated. METHODS: This paper considers data from three previous studies, each using a different breath sampling method. These include a peppermint washout study using continuous breath sampling with a purified air source, an exercise study using continuous breath sampling with an ambient air source, and a single breath sampling study with an ambient air source. For each dataset, different breath selection (data preprocessing) methods were compared and benchmarked according to predictive performance on a validation set and quantitative reliability of m/z bin intensity measurements. RESULTS: For both continuous methods, the best breath selection method improved the predictive model compared to no preselection, as measured by the 95% CI range for Youden's index, from 0.68-0.86 to 0.86-0.97 for the exercise study and 0.69-0.82 to 1.00-1.00 for the peppermint study. The reliability of intensity measurements for both datasets (as measured by median relative standard deviation (RSD)), was improved slightly by the best selection method compared to no preselection, from 18% to 14% for the exercise study and 7%-5% for the peppermint study. For the single breath samples, all the models resulted in perfect prediction, with a 95% CI range for Youden's index of 1.00-1.00. The reliability of the proposed method was 38%. CONCLUSION: The method of selecting exhaled breath from CMS data can affect the reliability of the measurement and the ability to distinguish between breath samples taken under different conditions. The application of appropriate data processing methods can improve the quality of the data and results obtained from CMS. The methods presented will enable untargeted analysis of breath VOCs using CMS to be performed.