Search

Clustering is an unsupervised machine learning methodology widely used in several sciences to find groups of similar patterns in complex data. The results generated by clustering algorithms generally depend on user-defined input parameters such as the number of expected clusters, which can have a great impact on the homogeneity of the identified clusters. Clustering validity indices (CVIs) are an effective method for determining the optimal number of clusters that best fit the natural partition of a dataset. They do not require any underlying assumption nor a priori knowledge about the true dataset structure. Since 1965, many cluster validity indices have been proposed in the literature and used in several different applications. In this paper, the performance of 68 cluster validity indices was evaluated on 21 real-life research and simulated datasets. CVIs were compared on the same partition for each dataset, which was searched for by the k-means clustering algorithm. Multivariate chemometric methods were applied to disclose mutual relationships among the indices and to select those that are more effective in terms of accuracy and reliability.

Approaches of high-level data fusion, also known as consensus, combine predictions of individual models to increase reliability and overcome limitations of single models. Consensus strategies are frequently applied in the framework of Quantitative Structure - Activity Relationships (QSARs) to reduce the uncertainties in the prediction of molecular activities and provide better accuracy of the model outcomes. However, specific regions of the chemical space may systematically be associated with low accuracy and even consensus modelling cannot improve prediction reliability through the multiple outcomes of individual models. In this study, a new heuristic metric to assess the degree of accuracy of consensus predictions in the chemical space is proposed. This metric can assist the mapping of reliability in prediction and enhance the delineation of a safe zone, where consensus predictions are expected to have better accuracy. The new metric is calculated by kernel-based potential functions and it can be used in the framework of both classification and regression consensus modelling. Four case studies, including extensive datasets for consensus modelling, were used to test the proposed approach. Results demonstrated that a potential can be associated with regions of the chemical space as a function of accuracy of consensus modelling and it can be used to enable the mapping of reliability in prediction and the definition of specific regions where predictions are expected to be more reliable.

The capacity to discriminate safe from dangerous compounds has played an important role in the evolution of species, including human beings. Highly evolved senses such as taste receptors allow humans to navigate and survive in the environment through information that arrives to the brain through electrical pulses. Specifically, taste receptors provide multiple bits of information about the substances that are introduced orally. These substances could be pleasant or not according to the taste responses that they trigger. Tastes have been classified into basic (sweet, bitter, umami, sour and salty) or non-basic (astringent, chilling, cooling, heating, pungent), while some compounds are considered as multitastes, taste modifiers or tasteless. Classification-based machine learning approaches are useful tools to develop predictive mathematical relationships in such a way as to predict the taste class of new molecules based on their chemical structure. This work reviews the history of multicriteria quantitative structure-taste relationship modelling, starting from the first ligand-based (LB) classifier proposed in 1980 by Lemont B. Kier and concluding with the most recent studies published in 2022.

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.

The chemometric philosophy is based on some well-defined and shared principles: multivariate analysis of the problem, searching for the relevant information of data, model validation to produce predictive models, definition of the model applicability domain for reliable predictions and graphical visualization of the relevant information. The verification of food authenticity is fundamentally a classification problem; indeed, samples are labeled according to their authenticity status, that is, authentic or non-authentic. Chemometric methods are usually applied to data that are collected in a numerical table. Chemometric methods usually require a pre-treatment of the raw data in order to be able to extract useful and non-trivial information and discard obvious (i.e., intrinsic in the single variable) and useless information. Soft independent modeling of class analogy is one of the most popular class modeling approaches. Artificial Neural Networks are currently considered among the most important emerging tools in chemometrics.

Mass spectrometry (MS) is widely used for the identification of chemical compounds by matching the experimentally acquired mass spectrum against a database of reference spectra. However, this approach suffers from a limited coverage of the existing databases causing a failure in the identification of a compound not present in the database. Among the computational approaches for mining metabolite structures based on MS data, one option is to predict molecular fingerprints from the mass spectra by means of chemometric strategies and then use them to screen compound libraries. This can be carried out by calibrating multi-task artificial neural networks from large datasets of mass spectra, used as inputs, and molecular fingerprints as outputs. In this study, we prepared a large LC-MS/MS dataset from an on-line open repository. These data were used to train and evaluate deep-learning-based approaches to predict molecular fingerprints and retrieve the structure of unknown compounds from their LC-MS/MS spectra. Effects of data sparseness and the impact of different strategies of data curing and dimensionality reduction on the output accuracy have been evaluated. Moreover, extensive diagnostics have been carried out to evaluate modelling advantages and drawbacks as a function of the explored chemical space.

The study concerns the photodegradation of the antidepressant escitalopram (ESC), the S-enantiomer of the citalopram raceme, both in ultrapure and surface water, considering the contribution of indirect photolysis through the presence of nitrate and bicarbonate. The effect of nitrate and bicarbonate concentrations was investigated by full factorial design, and only the nitrate concentration resulted in having a significant effect on the degradation. The kinetics of ESC photodegradation is the pseudo-first-order (half-life = 62.4 h in ultrapure water and 48.4 h in lake water). The generation of transformation products (TPs) was monitored through a developed and validated HPLC-MS/MS method. Fourteen TPs were identified in ultrapure water (one of them, at m/z 261, for the first time) and other two TPs at m/z 327 (found for the first time in this study) were identified only in presence of a nitrate. Several TPs were the same as those formed during the photodegradation of citalopram. The photodegradation pathway of ESC and its mechanism of degradation in water is proposed. The method was applied successfully to the analyses of surface water samples, in which a few dozen of ng L−1 of ESC was determined together with the presence of TP2, TP5 and TP12. Finally, a preliminary in silico evaluation of the toxicological profile and environmental behavior of TPs by computational models was carried out; two TPs (TP4 and TP10) were identified as of potential concern, as they were predicted mutagenic by Ames test model.

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 Squares-Discriminant Analysis (PLS-DA), K-Nearest Neighbour (KNN), exponentially weighted K-Nearest Neighbour (wKNN) and Support Vector Machine with radial functions (SVM-RBF) 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.