Neural networks applied to determine the thermophysical properties of amino acid based ionic liquids.

A series of models based on artificial neural networks (ANNs) have been designed to estimate the thermophysical properties of different amino acid-based ionic liquids (AAILs). Three different databases of AAILs were modeled using these algorithms with the goal set to estimate the density, viscosity, refractive index, ionic conductivity, and thermal expansion coefficient, and requiring only data regarding temperature and electronic polarizability of the chemicals. Additionally, a global model was designed combining all of the databases to determine the robustness of the method. In general, the results were successful, reaching mean prediction errors below 1% in many cases, as well as a statistically reliable and accurate global model. Attaining these successful models is a relevant fact as AAILs are novel biodegradable and biocompatible compounds which may soon make their way into the health sector forming a part of useful biomedical applications. Therefore, understanding the behavior and being able to estimate their thermophysical properties becomes crucial.

The accurate estimation of physicochemical properties of ternary mixtures containing ionic liquids via artificial neural networks.

The estimation of the density and refractive index of ternary mixtures comprising the ionic liquid (IL) 1-butyl-3-methylimidazolium tetrafluoroborate, 2-propanol, and water at a fixed temperature of 298.15 K has been attempted through artificial neural networks. The obtained results indicate that the selection of this mathematical approach was a well-suited option. The mean prediction errors obtained, after simulating with a dataset never involved in the training process of the model, were 0.050% and 0.227% for refractive index and density estimation, respectively. These accurate results, which have been attained only using the composition of the dissolutions (mass fractions), imply that, most likely, ternary mixtures similar to the one analyzed, can be easily evaluated utilizing this algorithmic tool. In addition, different chemical processes involving ILs can be monitored precisely, and furthermore, the purity of the compounds in the studied mixtures can be indirectly assessed thanks to the high accuracy of the model.

Artificial sensing intelligence with silicon nanowires for ultraselective detection in the gas phase.

The use of molecularly modified Si nanowire field effect transistors (SiNW FETs) for selective detection in the liquid phase has been successfully demonstrated. In contrast, selective detection of chemical species in the gas phase has been rather limited. In this paper, we show that the application of artificial intelligence on deliberately controlled SiNW FET device parameters can provide high selectivity toward specific volatile organic compounds (VOCs). The obtained selectivity allows identifying VOCs in both single-component and multicomponent environments as well as estimating the constituent VOC concentrations. The effect of the structural properties (functional group and/or chain length) of the molecular modifications on the accuracy of VOC detection is presented and discussed. The reported results have the potential to serve as a launching pad for the use of SiNW FET sensors in real-world counteracting conditions and/or applications.

Estimation of the refractive indices of imidazolium-based ionic liquids using their polarisability values.

Statistical models have been used to estimate the refractive index of 72 imidazolium-based ionic liquids using the electronic polarisability of their ions as the data for two different mathematical approaches: artificial neural networks, in the form of multi-layer perceptrons, and multiple linear regression models. Although the artificial neural networks and linear models have been able to accomplish this task, the multi-layer perceptron model has been shown to be a more accurate method, thanks to its ability of determining non-linear relationships between different dependent variables. Additionally, it is clear that the multiple linear regression presents a systematic deviation in the estimated refractive index values, which confirms that it is an inappropriate model for this system.

Deep quantification of a refined adulterant blended into pure avocado oil.

In this work, a new method has been developed to detect adulterations in avocado oil by combining optical images and their treatment with deep learning algorithms. For this purpose, samples of avocado oil adulterated with refined olive oil at concentrations from 1 % to 15 % (v/v) were prepared. Two groups of images of the different samples were obtained, one in conditions considered as bright and the other as dark, obtaining a total of 1,800 photographs. To obtain these images under both conditions, the exposure or shutter speed of the camera was modified (1/30 s for light conditions and 1/500 s for dark conditions). A residual neural network (ResNet34) was used to process and classify the images obtained. A different model was developed for each condition, and during blind validation of the models, ∼95 % of the images were correctly classified.

Is my food safe? - AI-based classification of lentil flour samples with trace levels of gluten or nuts.

An artificial intelligence-based method to rapidly detect adulterated lentil flour in real time is presented. Mathematical models based on convolutional neural networks and transfer learning (viz., ResNet34) have been trained to identify lentil flour samples that contain trace levels of wheat (gluten) or pistachios (nuts), aiding two relevant populations (people with celiac disease and with nut allergies, respectively). The technique is based on the analysis of photographs taken by a simple reflex camera and further classification into groups assigned to adulterant type and amount (up to 50 ppm). Two different algorithms were trained, one per adulterant, using a total of 2200 images for each neural network. Using blind sets of data (10% of the collected images; initially and randomly separated) to evaluate the performance of the models led to strong performances, as 99.1% of lentil flour samples containing ground pistachio were correctly classified, while 96.4% accuracy was reached to classify the samples containing wheat flour.

Convolutional capture of the expansion of extra virgin olive oil droplets to quantify adulteration.

In this research, more than 302,000 images of five different types of extra virgin olive oils (EVOOs) have been collected to train and validate a system based on convolutional neural networks (CNNs) to carry out their classification. Furthermore, comparable deep learning models have also been trained to detect and quantify the adulteration of these EVOOs with other vegetable oils. In this work, three groups of CNN models have been tested for (i) the classification of all EVOOs, (ii) the detection and quantification of adulterated samples for each individual EVOO, and (iii) a global version of the previous models combining all EVOOs into a single quantifying CNN. This last model was successfully validated using 30,195 images that were initially isolated from the initial database. The result was an algorithm capable of detecting and accurately classifying the five types of EVOO and their respective adulteration concentrations with an overall hit rate of >96%. Therefore, EVOO droplet analyses via CNNs have proven to be a convincing quality control tool for the evaluation of EVOO, which can be carried by producers, distributors, or even final consumers, to help locate adulterations.

Distinct thermal patterns to detect and quantify trace levels of wheat flour mixed into ground chickpeas.

This paper combines intelligent algorithms based on a residual neural network (ResNet34) to process thermographic images. This integration is aimed at detecting traces of wheat flour, in concentrations from 1 to 50 ppm, mixed into chickpea flour. Using an image database of over 16 thousand samples to train the ResNet34, and 1712 images to blindly test it, the optimized intelligent algorithm is able to classify the thermographic images into 14 classes according to the concentration of wheat flour at a 99.0% correct classification rate. These results open the door to the development of a simple, fast, and inexpensive prototype that can be used during the entire distribution chain to help protect brands and consumers. The detection and quantification of trace amounts of wheat flour, or indirectly gluten, serves as a quality control and health safety application protecting, for example, people with celiac disease.

Prediction of gas solubilities in ionic liquids.

Ionic liquids (of which it is estimated that there are at least one million simple fluids) generate a rich chemical space, which is now just at the beginning of its systematic exploration. Many properties of ionic liquids are truly unique and, which is more important, can be finely tuned. Differential solubility of industrial chemicals in ionic liquids is particularly interesting, because it can be a basis for novel, efficient, environmentally friendly technologies. Given the vast number of potential ionic liquids, and the impossibility of a comprehensive empirical exploration, it is essential to extract the maximum information from extant data. We report here some computational models of gas solubility. These multiple regression- and neural network-based models cover a chemical space spanned by 48 ionic liquids and 23 industrially important gases. Molecular polarisabilities and special Lewis acidity and basicity descriptors calculated for the ionic liquid cations and anions, as well as for the gaseous solutes, are used as input parameters. The quality of fit "observed versus predicted Henry's law constants" is particularly good for the neural network model. Validation was established with an external dataset, again with a high quality fit. In contrast to many other neural network models published, our model is no "black box", since contributions of the parameters and their nonlinearity characteristics are calculated and analysed.

Profiles of Volatile Biomarkers Detect Tuberculosis from Skin.

Tuberculosis (TB) is an infectious disease that threatens >10 million people annually. Despite advances in TB diagnostics, patients continue to receive an insufficient diagnosis as TB symptoms are not specific. Many existing biodiagnostic tests are slow, have low clinical performance, and can be unsuitable for resource-limited settings. According to the World Health Organization (WHO), a rapid, sputum-free, and cost-effective triage test for real-time detection of TB is urgently needed. This article reports on a new diagnostic pathway enabling a noninvasive, fast, and highly accurate way of detecting TB. The approach relies on TB-specific volatile organic compounds (VOCs) that are detected and quantified from the skin headspace. A specifically designed nanomaterial-based sensors array translates these findings into a point-of-care diagnosis by discriminating between active pulmonary TB patients and controls with sensitivity above 90%. This fulfills the WHO's triage test requirements and poses the potential to become a TB triage test.

A COSMO-RS based guide to analyze/quantify the polarity of ionic liquids and their mixtures with organic cosolvents.

A COSMO-RS descriptor (S(sigma-profile)) has been used in quantitative structure-property relationship (QSPR) studies by a neural network (NN) for the prediction of empirical solvent polarity E(T)(N) scale of neat ionic liquids (ILs) and their mixtures with organic solvents. S(sigma-profile) is a two-dimensional quantum chemical parameter which quantifies the polar electronic charge of chemical structures on the polarity (sigma) scale. Firstly, a radial basis neural network exact fit (RBNN) is successfully optimized for the prediction of E(T)(N), the solvatochromic parameter of a wide variety of neat organic solvents and ILs, including imidazolium, pyridinium, ammonium, phosphonium and pyrrolidinium families, solely using the S(sigma-profile) of individual molecules and ions. Subsequently, a quantitative structure-activity map (QSAM), a new concept recently developed, is proposed as a valuable tool for the molecular understanding of IL polarity, by relating the E(T)(N) polarity parameter to the electronic structure of cations and anions given by quantum-chemical COSMO-RS calculations. Finally, based on the additive character of the S(sigma-profile) descriptor, we propose to simulate the mixture of IL-organic solvents by the estimation of the S(sigma-profile)(Mixture) descriptor, defined as the weighted mean of the S(sigma-profile) values of the components. Then, the E(T)(N) parameters for binary solvent mixtures, including ILs, are accurately predicted using the S(sigma-profile)(Mixture) values from the RBNN model previously developed for pure solvents. As result, we obtain a unique neural network tool to simulate, with similar reliability, the E(T)(N) polarity of a wide variety of pure ILs as well as their mixtures with organic solvents, which exhibit significant positive and negative deviations from ideality.

Machine Learning and Feature Selection Applied to SEER Data to Reliably Assess Thyroid Cancer Prognosis.

Utilizing historical clinical datasets to guide future treatment choices is beneficial for patients and physicians. Machine learning and feature selection algorithms (namely, Fisher's discriminant ratio, Kruskal-Wallis' analysis, and Relief-F) have been combined in this research to analyse a SEER database containing clinical features from de-identified thyroid cancer patients. The data covered 34 unique clinical variables such as patients' age at diagnosis or information regarding lymph nodes, which were employed to build various novel classifiers to distinguish patients that lived for over 10 years since diagnosis, from those who did not survive at least five years. By properly optimizing supervised neural networks, specifically multilayer perceptrons, using data from large groups of thyroid cancer patients (between 6,756 and 20,344 for different models), we demonstrate that unspecialized and existing medical recording can be reliably turned into power of prediction to help doctors make informed and optimized treatment decisions, as distinguishing patients in terms of prognosis has been achieved with 94.5% accuracy. We also envisage the potential of applying our machine learning strategy to other diseases and purposes such as in designing clinical trials for unmasking the maximum benefits and minimizing risks associated with new drug candidates on given populations.

Convolutional decoding of thermographic images to locate and quantify honey adulterations.

In this research, 56 samples of pure honey have been mixed with different concentrations of rice syrup simulating a set of adulterated samples. A thermographic camera was used to extract data regarding the thermal development of the honey. The resulting infrared images were processed via convolutional neural networks (CNNs), a subset of algorithms within deep learning. The CNNs have been trained and optimized using these images to detect the commonly elusive rice syrup in honey in concentrations as low as 1% in weight, as well as quantify it. Finally, the model was successfully validated using images which were initially isolated from the training database. The result was an algorithm capable of identifying adulterated honey from different floral origins and quantifying rice syrup with accuracies of 95% and 93%, respectively. Therefore, CNNs have complemented the thermographic analysis and have shown to be a compelling tool for the control of food quality, thanks to traits such as high sensitivity, speed, and being independent of highly specialized personnel.

Chaotic parameters from fluorescence spectra to resolve fraudulent mixtures of fresh and expired protected designation of origin extra virgin olive oils.

One of the most profitable products from the Mediterranean basin is extra virgin olive oil (EVOO), and, therefore, some of them have protected designation of origin (PDO) labels. In order to prevent fraudulent practices, a method to quantify adulterants has been developed. 459 binary blends composed of PDO EVOO in date (Saqura, Oleoestepa, and Duque de Baena) mixed with expired PDO EVOO (Quinta do Vallouto, Señorío de Segura, and Planeta) to serve as adulterants (<17%) have been analyzed. Using a laser diode as a source light, the fluorescence emission has been measured and 20 chaotic parameters from the resulting spectra have been calculated. Using these as independent variables of multi-parameter regression models, the concentration of adulterant has been estimated. Every model was evaluated through the mean square error, adjusted correlation coefficient, Mallows' Cp, Akaike information criterion, Hannan-Quinn criterion, and Bayesian information criterion. This approach was validated by the leave-one-out cross-validation method and the results were promising (lower than 10% quantification error). Additionally, the structure of the sensor has been designed and developed by a 3D printer and has the potential of being applied in situ for real-time and cost-effective analysis at oil mills or for quality control.

Combination of LEDs and cognitive modeling to quantify sheep cheese whey in watercourses.

The concentration of sheep cheese whey (CW) in water obtained from two Spanish reservoirs, two Spanish rivers, and distilled water has been estimated by combining spectroscopic measurements, obtained with light-emitting diodes (LEDs), and linear or non-linear algorithms. The concentration range of CW that has been studied covers from 0 to 25% in weight. Every sample was measured by six different types of LEDs possessing different emission wavelengths (blue, orange, green, pink, white, and UV). 1,800 fluorescence measurements were carried out and used to design different types of models to estimate the concentration of CW in water. The fluorescence spectra provided by the pink LED originated the most accurate mathematical models, with mean square errors lower than 3.3% and 2.5% for the linear and non-linear approaches, respectively. The pink LED combined with the non-linear model, which was an artificial neural network, was further validated through a k-fold cross-validation and an internal validation. It should be noted that the sensor used here has been designed and produced by a 3D printer and has the potential of being implemented in situ for real-time and cost-effective analysis of natural watercourses.

Chaotic parameters extracted from fluorescence spectra to quantify sheep cheese whey in natural bodies of water.

Sheep cheese whey (SCW) is a by-product from the dairy industry, and due to its composition, it is very hazardous for natural bodies of water. However, illegal discharges of this product have been commonly reported in watercourses and reservoirs. To prevent this type of actions, a simple and affordable sensor has been designed and validated using diverse water samples from different sources containing SCW, such as water from two Spanish reservoirs and two Spanish rivers located in the province of Madrid. Using these waters, different SCW solutions (lower than 20% in weight) have been prepared and measured. The equipment used to sense the samples is based on combining fluorescence measurements, obtained with light emitting diodes (LEDs), and algorithms which rely on chaotic parameters. Every sample was measured by six different types of LEDs possessing distinct emission wavelengths (blue, orange, green, pink, white, and UV), leading to 1786 fluorescence spectra that were employed during the modeling phase. After the mathematical analysis, the dataset that generates the best statistical results was from the blue LED. This approach was dually validated via leave-one-out cross-validation as well as externally validation, and the results were very promising (error around 6.5% and 8% quantification error, respectively). Additionally, it is important to note that the sensor used has been designed and developed by a 3D printer and has the potential of being applied in situ for real-time and cost-effective analysis of natural bodies of water.

Characterization of an array of honeys of different types and botanical origins through fluorescence emission based on LEDs.

A set of 10 honeys comprising a diverse range of botanical origins have been successfully characterized through fluorescence spectroscopy using inexpensive light-emitting diodes (LEDs) as light sources. It has been proven that each LED-honey combination tested originates a unique emission spectrum, which enables the authentication of every honey, being able to correctly label it with its botanical origin. Furthermore, the analysis was backed up by a mathematical analysis based on partial least square models which led to a correct classification rate of each type of honey of over 95%. Finally, the same approach was followed to analyze rice syrup, which is a common honey adulterant that is challenging to identify when mixed with honey. A LED-dependent and unique fluorescence spectrum was found for the syrup, which presumably qualifies this approach for the design of uncomplicated, fast, and cost-effective quality control and adulteration assessing tools for different types of honey.

Quantification of phenolic compounds in olive oil mill wastewater by artificial neural network/laccase biosensor.

In this paper is considered a new computerized approach to the determination of concentrations of phenolic compounds (caffeic acid and catechol). An integrated artificial neural network (ANN)/laccase biosensor is designed. The data collected (current signals) from amperometric detection of the laccase biosensor were transferred into an ANN trained computer for modeling and prediction of output. Such an integrated ANN/laccase biosensor system is capable of the prediction of caffeic acid and catechol concentrations of olive oil mill wastewater, based on the created models and patterns, without any previous knowledge of this phenomenon. The predicted results using the ANN were compared with the amperometric detection of phenolic compounds obtained at a laccase biosensor in olive oil wastewater of the 2004-2005 harvest season. The difference between the real and the predicted values was <0.5%. biosensor; olive oil mill wastewater; chemical analysis; phenolic compounds.

Cancer metabolism: the volatile signature of glycolysis-in vitro model in lung cancer cells.

Discovering the volatile signature of cancer cells is an emerging approach in cancer research, as it may contribute to a fast and simple diagnosis of tumors in vivo and in vitro. One of the main contributors to such a volatile signature is hyperglycolysis, which characterizes the cancerous cell. The metabolic perturbation in cancer cells is known as the Warburg effect; glycolysis is preferred over oxidative phosphorylation (OXPHOS), even in the presence of oxygen. The precise mitochondrial alterations that underlie the increased dependence of cancer cells on aerobic glycolysis for energy generation have remained a mystery. We aimed to profile the volatile signature of the glycolysis activity in lung cancer cells. For that an in vitro model, using lung cancer cell line cultures (A549, H2030, H358, H322), was developed. The volatile signature was measured by proton transfer reaction mass spectrometry under normal conditions and glycolysis inhibition. Glycolysis inhibition and mitochondrial activity were also assessed by mitochondrial respiration capacity measurements. Cells were divided into two groups upon their glycolytic profile (PET positive and PET negative). Glycolysis blockade had a unique characteristic that was shared by all cells. Furthermore, each group had a characteristic volatile signature that enabled us to discriminate between those sub-groups of cells. In conclusion, lung cancer cells may have different subpopulations of cells upon low and high mitochondrial capacity. In both groups, glycolysis blockade induced a unique volatile signature.

Exhaled Breath Markers for Nonimaging and Noninvasive Measures for Detection of Multiple Sclerosis.

Multiple sclerosis (MS) is the most common chronic neurological disease affecting young adults. MS diagnosis is based on clinical characteristics and confirmed by examination of the cerebrospinal fluids (CSF) or by magnetic resonance imaging (MRI) of the brain or spinal cord or both. However, neither of the current diagnostic procedures are adequate as a routine tool to determine disease state. Thus, diagnostic biomarkers are needed. In the current study, a novel approach that could meet these expectations is presented. The approach is based on noninvasive analysis of volatile organic compounds (VOCs) in breath. Exhaled breath was collected from 204 participants, 146 MS and 58 healthy control individuals. Analysis was performed by gas-chromatography mass-spectrometry (GC-MS) and nanomaterial-based sensor array. Predictive models were derived from the sensors, using artificial neural networks (ANNs). GC-MS analysis revealed significant differences in VOC abundance between MS patients and controls. Sensor data analysis on training sets was able to discriminate in binary comparisons between MS patients and controls with accuracies up to 90%. Blinded sets showed 95% positive predictive value (PPV) between MS-remission and control, 100% sensitivity with 100% negative predictive value (NPV) between MS not-treated (NT) and control, and 86% NPV between relapse and control. Possible links between VOC biomarkers and the MS pathogenesis were established. Preliminary results suggest the applicability of a new nanotechnology-based method for MS diagnostics.