Search

The taxonomy of calcareous red algae continues to undergo revisions, especially since the spread of molecular techniques (Pezzolesi et al., 2019; Caragnano et al., 2021). Genetic analyses challenged the morphological approach to species identification, that relies on thallus organization, morphology of cells and reproductive structures observed with high magnification scanning electron microscopy (SEM), or in thin sections under optical microscope. Recently, a growing attention is also given to the calcified structures and to the shape of the crystallites composing the cell walls, which are diagnostic at the level of family (Auer and Piller, 2020). Looking for new identification tools therefore represents a current challenge, which could have crucial implications in paleontology. Convolutional neural networks (CNN) have successfully been applied for the classification of fish, bivalves, and foraminifera (de Lima et al., 2020). To the best of our knowledge, there is no literature about CNN for species identification using SEM images. We tested the potential of CNN to classify SEM images taken from different species of coralline algae commonly found in Mediterranean waters: Lithothamnion corallioides (P. Crouan & H. Crouan) P. Crouan & H. Crouan 1867, Mesophyllum philippii (Foslie) Adey 1970 and Lithophyllum racemus (Lamarck) Foslie 1901. The model provided promising results in terms of accuracy, considering the small set of images used (~40 per species). SEM acquisition, indeed, has significant costs that limit the number of images available for cross-validation. Further efforts should focus on enhancing the image dataset to improve the CNN classification capability and on the interpretation of the features determining the classes.

Liquid chromatography tandem mass spectrometry (LC-MS/MS) is routinely used in many clinical applications, including toxicology, drug monitoring, endocrinology, microbiology, and proteomics, thanks to its versatility and effectiveness in the determination of small molecules. The standard approach for analyzing spectral data is the spectral matching that is based on creating a library of annotated spectra against which individual spectrum can be searched for. Building an own spectral library is time-consuming and dependent on the LC-MS/MS instrumentation used, whereas freely available libraries only cover a limited number of molecules. In many cases and especially with newly synthetized substances, the spectrum could not be found in any spectral databases. In-silico fingerprints are binary vectors that, through a hashing algorithm, encode features of molecules. Due to their easy computation and instrument-independence, in-silico fingerprint databases are bigger and more updated than spectral ones. Prediction of molecular fingerprints starting from the LC-MS/MS spectra would consequently assist the match of any target compound under investigation, which would benefit from the increased dimension of fingerprint databases. In this study, we developed a multi-task neural network able to predict molecular fingerprints starting from the LC-MS/MS spectra. We initially collected and pruned around 70’000 MS spectra from available sources (MassBank of North America). For each compound, fingerprints were calculated (MACCS167 keys and Dragon ECFP of 512 bits) and then multi-task feedforward neural networks were trained to predict the binary bits of molecular fingerprints. Models were validated through specific validation protocols and demonstrated to have suitable performances in terms of predictive accuracy (>85% of bits correctly predicted). Thus, the proposed networks can represent a potential approach for the development of a reliable method to enhance matching of MS spectral data to larger molecular databases. Moreover, with respect to recent studies, the proposed model is simple and works with spectra obtained in different experimental conditions.

Quantifying the diversity content of an incidence matrix is challenging in several scientific fields. The existing indices capture diverse facets of diversity and thus comparing their behaviour is not a straightforward task. For example, an application of diversity measures involves ensembles of classifiers which usually in real applications contain missing values. Therefore, we analysed 14 statistics and, after making them comparable and able to deal with missing values, we applied them on more than one hundred incidence matrices in order to examine the relationships among the measures themselves. In particular, we highlighted the importance of the inter-row agreement of factors, the general agreement of incident factors, as well as the influence on the indices of the proportion of missing values and matrix dimensions, the sensitivity to missing values, the uniform distribution of entries and the invariance to matrix transposition.

In the last ten years, the concept of activity cliff gained a lot of interest by medicinal chemists and QSAR modellers. The reason of this interest is its capacity to highlight problematic as well as interesting situations in analysing the activity/property of pairs of compounds, thus focusing the research on the reasons why two similar compounds show different activities. In the literature, activity cliffs as well as their dual aspect, i.e. structural cliffs, are calculated from activities represented by a continuous response. In this paper, the proposal is to estimate the activity cliffs of categorical responses, i.e. for compounds partitioned into different classes.

Multitask learning allows to model multiple tasks simultaneously through information sharing. In the context of quantitative structure–activity relationships and computational toxicology, multitask learning is gaining more and more interest, owed to its potential to improve the predictive performance of underrepresented tasks and to predict the multi-property profile of molecules. In this chapter, after introducing the multitask problem formulation, we present a hands-on tutorial on multitask neural networks.

I recettori nucleari sono fattori di trascrizione coinvolti in processi critici per la salute umana e sono un obiettivo rilevante per la valutazione del rischio tossicologico e il processo di scoperta dei farmaci. I modelli computazionali possono essere uno strumento utile (i) per dare priorità alla sperimentazione di sostanze chimiche che possono imitare gli ormoni naturali e quindi essere interferenti endocrini e (ii) per identificare nuovi possibili candidati farmaci. Pertanto, l'obiettivo principale di questo progetto è quello di studiare le potenziali interazioni tra sostanze chimiche e recettori nucleari, con il duplice scopo di sviluppare strumenti in silico per la ricerca di nuovi modulatori e di identificare possibili sostanze chimiche che alterano il sistema endocrino. Dopo aver creato una collezione esaustiva di modulatori di recettori nucleari, abbiamo applicato metodi di apprendimento automatico per colmare il vuoto di dati e predire nuovi possibili modulatori tramite modelli predittivi. In particolare, le strategie di modellazione hanno incluso algoritmi di apprendimento automatico multi-tasking per indagare le complesse relazioni tra sostanze chimiche e diversi recettori nucleari. Nuclear receptors are transcription factors involved in processes critical to human health and are a relevant target for toxicological risk assessment and the drug discovery process. Computational models can be a useful tool (i) to prioritize chemicals that can mimic natural hormones and thus be endocrine disruptors and (ii) to identify new possible lead for drug discovery. Therefore, the main goal of this project is to study potential interactions between chemicals and nuclear receptors, with the dual purpose of developing in silico tools to search for new modulators and to identify possible endocrine disrupting chemicals. After creating an exhaustive collection of nuclear receptor modulators, we applied machine learning methods to fill the data gap and prioritize modulators by building predictive models. In particular, modeling strategies included multi-tasking machine learning algorithms to investigate the complex relationships between chemicals and multiple nuclear receptors.

Minimum Spanning Tree (MST) is a well-known clustering algorithm that provides a graphical tree representation of the objects in a data set by exploiting local information to link each pair of similar objects. The a-posteriori analysis of this tree in terms of nodes and edges provides the basis to derive simple classifiers, namely semi-supervised classification approaches based on the minimum spanning tree approach. In this work, we propose different metrics to evaluate the MST ability to group objects of the same a-priori known classes. The classification capability of the proposed approach, using 13 different distance measures, was compared with that of classical supervised classification approaches such as N-Nearest Neighbour (N3), Binned Nearest Neighbour (BNN), Partial Least SquaresDiscriminant Analysis (PLS-DA), K-Nearest Neighbour (KNN), exponentially weighted K-Nearest Neighbour (wKNN) and Support Vector Machine with radial functions (SVMRBF) on 31 data sets. The proposed approach resulted to be competitive and comparable with the considered classical supervised classification methods. Finally, we analysed the role of the 13 different measures in terms of performance and percentage of not-assigned objects.

Antineoplastic drugs are a wide and heterogeneous group of substances that, as universally known, can cause highly severe toxic effects to whoever is exposed. From an occupational safety point of view, surface contaminations inside preparation and administration units are a growing issue and therefore require the development and implementation of sensible and fast monitoring methods. The unlikelihood of a unique all-embracing chromatography, able to correctly retain and separate each analyte led to the need to create an orthogonal normal phase analysis, which might be able to fill the gaps in the more common reversed-phase ones. An existing hydrophilic interaction method has thus been expanded to 6 other drugs and applied to real samples after an evaluation of its performances. The experimental data were then used to evaluate the possibility of estimating reliable relationships between the chromatographic retention and the chemical-structural features of the drugs under analysis.

The interest in multitask and deep learning strategies has been increasing in the last few years, in application to large and complex dataset for quantitative structure-activity relationship (QSAR) analysis. Multitask approaches allow the simultaneous prediction of molecular properties that are related, through information sharing, whereas deep learning strategies increase the potential of capturing nonlinear relationships. In this work, we compare the binary classification capability of multitask deep and shallow neural networks to single-task strategies used as benchmark (i.e., as k-nearest neighbours, N-nearest neighbours, random forest and Naïve Bayes), as well as multitask supervised self-organizing maps. Comparison was carried out with an extended QSAR dataset containing annotations of molecular binding, agonism and antagonism activity on 11 nuclear receptors, for a total of 14,963 molecules, divided into training and test sets and labelled for their bioactivity on at least one of 30 binary tasks. Additional 304 chemicals were used as external evaluation set to further validate models. Although no approach systematically overperformed the others, task-specific differences were found, suggesting the benefit of multitask learning for tasks that are less represented. On average, some of the single-task approaches and multitask deep learning strategies had similar performances. However, the latter can have advantages, such as a simpler management of predictions and applicability domain assessment for future samples. On the other hand, the parameter tuning required by neural networks are generally time expensive suggesting that the modelling strategy should be evaluated case by case.

The classification of coralline algae commonly relies on the morphology of cells and reproductive structures, along with thallus organization, observed through Scanning Electron Microscopy (SEM). Nevertheless, species identification based on morphology often leads to uncertainty, due to their general plasticity. Evolutionary and environmental studies featured coralline algae for their ecological significance in both recent and past Oceans and need to rely on robust taxonomy. Research efforts towards new putative diagnostic tools have recently been focused on cell wall ultrastructure. In this work, we explored a new classification tool for coralline algae, using fine-tuning pretrained Convolutional Neural Networks (CNNs) on SEM images paired to morphological categories, including cell wall ultrastructure. We considered four common Mediterranean species, classified at genus and at the species level (Lithothamnion corallioides, Mesophyllum philippii, Lithophyllum racemus, Lithophyllum pseudoracemus). Our model produced promising results in terms of image classification accuracy given the constraint of a limited dataset and was tested for the identification of two ambiguous samples referred to as L. cf. racemus. Overall, explanatory image analyses suggest a high diagnostic value of calcification patterns, which significantly contributed to class predictions. Thus, CNNs proved to be a valid support to the morphological approach to taxonomy in coralline algae.

During the COVID-19 pandemic, a shortage of mechanical ventilators was reported and ventilator sharing between patients was proposed as an ultimate solution. Two lung simulators were ventilated by one anesthesia machine connected through two respiratory circuits and T-pieces. Five different combinations of compliances (30–50 mL × cmH2O−1) and resistances (5–20 cmH2O × L−1 × s−1) were tested. The ventilation setting was: pressure-controlled ventilation, positive end-expiratory pressure 15 cmH2O, inspiratory pressure 10 cmH2O, respiratory rate 20 bpm. Pressures and flows from all the circuit sections have been recorded and analyzed. Simulated patients with equal compliance and resistance received similar ventilation. Compliance reduction from 50 to 30 mL × cmH2O−1 decreased the tidal volume (VT) by 32% (418 ± 49 vs. 285 ± 17 mL). The resistance increase from 5 to 20 cmH2O × L−1 × s−1 decreased VT by 22% (425 ± 69 vs. 331 ± 51 mL). The maximal alveolar pressure was lower at higher compliance and resistance values and decreased linearly with the time constant (r² = 0.80, p &lt, 0.001). The minimum alveolar pressure ranged from 15.5 ± 0.04 to 16.57 ± 0.04 cmH2O. Cross-flows between the simulated patients have been recorded in all the tested combinations, during both the inspiratory and expiratory phases. The simultaneous ventilation of two patients with one ventilator may be unable to match individual patient’s needs and has a high risk of cross-interference.