Search

Natural products are a diverse class of compounds with promising biological properties, such as high potency and excellent selectivity. However, they have different structural motifs than typical drug-like compounds, e.g., a wider range of molecular weight, multiple stereocenters and higher fraction of sp3-hybridized carbons. This makes the encoding of natural products via molecular fingerprints difficult, thus restricting their use in cheminformatics studies. To tackle this issue, we explored over 30 years of research to systematically evaluate which molecular fingerprint provides the best performance on the natural product chemical space. We considered 20 molecular fingerprints from four different sources, which we then benchmarked on over 100,000 unique natural products from the COCONUT (COlleCtion of Open Natural prodUcTs) and CMNPD (Comprehensive Marine Natural Products Database) databases. Our analysis focused on the correlation between different fingerprints and their classification performance on 12 bioactivity prediction datasets. Our results show that different encodings can provide fundamentally different views of the natural product chemical space, leading to substantial differences in pairwise similarity and performance. While Extended Connectivity Fingerprints are the de-facto option to encoding drug-like compounds, other fingerprints resulted to match or outperform them for bioactivity prediction of natural products. These results highlight the need to evaluate multiple fingerprinting algorithms for optimal performance and suggest new areas of research. Finally, we provide an open-source Python package for computing all molecular fingerprints considered in the study, as well as data and scripts necessary to reproduce the results, at https://github.com/dahvida/NP_Fingerprints.

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.

Artificial intelligence (AI) in the form of deep learning has promise for drug discovery and chemical biology, for example, to predict protein structure and molecular bioactivity, plan organic synthesis, and design molecules de novo. While most of the deep learning efforts in drug discovery have focused on ligand-based approaches, structure-based drug discovery has the potential to tackle unsolved challenges, such as affinity prediction for unexplored protein targets, binding-mechanism elucidation, and the rationalization of related chemical kinetic properties. Advances in deep-learning methodologies and the availability of accurate predictions for protein tertiary structure advocate for a renaissance in structure-based approaches for drug discovery guided by AI. This review summarizes the most prominent algorithmic concepts in structure-based deep learning for drug discovery, and forecasts opportunities, applications, and challenges ahead.

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.

Background: Humans are exposed to tens of thousands of chemical substances that need to be assessed for their potential toxicity. Acute systemic toxicity testing serves as the basis for regulatory hazard classification, labeling, and risk management. However, it is cost- and time-prohibitive to evaluate all new and existing chemicals using traditional rodent acute toxicity tests. In silica models built using existing data facilitate rapid acute tox- icity predictions without using animals. Objkctivks: The U.S. Interagency Coordinating Committee on the Validation of Alternative Methods (ICCVAM) Acute Toxicity Workgroup organ- ized an international collaboration to develop in silica models for predicting acute oral toxicity based on five different end points: Lethal Dose 50 (LD50 value, U.S. Environmental Protection Agency hazard (four) categories, Globally Harmonized System for Classification and Labeling hazard (five) categories, very toxic chemicals [LD50 (LD50 ≤ 50 mg/kg)], and nontoxic chemicals (LD50 > 2,000 mg/kg). Mkthods: An acute oral toxicity data inventory for 11,992 chemicals was compiled, split into training and evaluation sets, and made available to 35 participating international research groups that submitted a total of 139 predictive models. Predictions that fell within the applicability domains of the submitted models were evaluated using external validation sets. These were then combined into consensus models to leverage strengths of individual approaches. Results: The resulting consensus predictions, which leverage the collective strengths of each individual model, form the Collaborative Acute Toxicity Modeling Suite (CATMoS). CATMoS demonstrated high performance in terms of accuracy and robustness when compared with in viva results. Discussion: CATMoS is being evaluated by regulatory agencies for its utility and applicability as a potential replacement for in viva rat acute oral toxicity studies. CATMoS predictions for more than 800,000 chemica

Consensus strategies have been widely applied in many different scientific fields, based on the assumption that the fusion of several sources of information increases the outcome reliability. Despite the widespread application of consensus approaches, their advantages in quantitative structure-activity relationship (QSAR) modeling have not been thoroughly evaluated, mainly due to the lack of appropriate large-scale data sets. In this study, we evaluated the advantages and drawbacks of consensus approaches compared to single classification QSAR models. To this end, we used a data set of three properties (androgen receptor binding, agonism, and antagonism) for approximately 4000 molecules with predictions performed by more than 20 QSAR models, made available in a large-scale collaborative project. The individual QSAR models were compared with two consensus approaches, majority voting and the Bayes consensus with discrete probability distributions, in both protective and nonprotective forms. Consensus strategies proved to be more accurate and to better cover the analyzed chemical space than individual QSARs on average, thus motivating their widespread application for property prediction. Scripts and data to reproduce the results of this study are available for download.

BACKGROUND: Endocrine disrupting chemicals (EDCs) are xenobiotics that mimic the interaction of natural hormones and alter synthesis, transport, or metabolic pathways. The prospect of EDCs causing adverse health effects in humans and wildlife has led to the development of scientific and regulatory approaches for evaluating bioactivity. This need is being addressed using high-throughput screening (HTS) in vitro approaches and computational modeling. OBJECTIVES: In support of the Endocrine Disruptor Screening Program, the U.S. Environmental Protection Agency (EPA) led two worldwide consortiums to virtually screen chemicals for their potential estrogenic and androgenic activities. Here, we describe the Collaborative Modeling Project for Androgen Receptor Activity (CoMPARA) efforts, which follows the steps of the Collaborative Estrogen Receptor Activity Prediction Project (CERAPP). METHODS: The CoMPARA list of screened chemicals built on CERAPP’s list of 32,464 chemicals to include additional chemicals of interest, as well as simulated ToxCastTM metabolites, totaling 55,450 chemical structures. Computational toxicology scientists from 25 international groups contributed 91 predictive models for binding, agonist, and antagonist activity predictions. Models were underpinned by a common training set of 1,746 chemicals compiled from a combined data set of 11 ToxCastTM/Tox21 HTS in vitro assays. RESULTS: The resulting models were evaluated using curated literature data extracted from different sources. To overcome the limitations of single-model approaches, CoMPARA predictions were combined into consensus models that provided averaged predictive accuracy of approximately 80% for the evaluation set. DISCUSSION: The strengths and limitations of the consensus predictions were discussed with example chemicals; then, the models were implemented into the free and open-source OPERA application to enable screening of new chemicals with a defined applicability domain and accuracy ass

Chemometrics plays a fundamental role in quantitative structure-activity relationships (QSARs) and quantitative structure-property relationships (QSPRs) methods, which aim at empirically linking the molecular structure of chemicals to experimentally-measurable properties. In fact, chemometrics statistics and chemoinformatics are the basic tools for finding meaningful mathematical relationships between the molecular structure and biological, physico-chemical, toxicological and environmental properties of chemicals. The key elements of QSAR/QSPR are molecular descriptors, which are numerical indices encoding information related to the structure of chemicals and are used as independent variables in the subsequent modeling, thus connecting the QSAR approaches to the multivariate and chemometric world. In this article, historical and novel QSAR modeling approaches are presented. After describing the historical background of QSAR and the classical approaches, molecular descriptors are illustrated, with state-of-the-art as well as novel description methodologies. Additionally, the key principles of QSAR are introduced, along with specific elements of the QSAR modeling workflow, e.g., variable reduction and selection, similarity-based approaches, validation, the definition of applicability domain and consensus modeling.

Nuclear receptors (NRs) are key regulators of human health and constitute a relevant target for medicinal chemistry applications as well as for toxicological risk assessment. Several open databases dedicated to small molecules that modulate NRs exist; however, depending on their final aim (i.e., adverse effect assessment or drug design), these databases contain a different amount and type of annotated molecules, along with a different distribution of experimental bioactivity values. Stemming from these considerations, in this work we aim to provide a unified dataset, NURA (NUclear Receptor Activity) dataset, collecting curated information on small molecules that modulate NRs, to be intended for both pharmacological and toxicological applications. NURA contains bioactivity annotations for 15,247 molecules and 11 selected NRs, and it was obtained by integrating and curating data from toxicological and pharmacological databases (i.e., Tox21, ChEMBL, NR-DBIND and BindingDB). Our results show that NURA dataset is a useful tool to bridge the gap between toxicology- and medicinal-chemistry-related databases, as it is enriched in terms of number of molecules, structural diversity and covered atomic scaffolds compared to the single sources. To the best of our knowledge, NURA dataset is the most exhaustive collection of small molecules annotated for their modulation of the chosen nuclear receptors. NURA dataset is intended to support decision-making in pharmacology and toxicology, as well as to contribute to data-driven applications, such as machine learning. The dataset and the data curation pipeline can be downloaded free of charge on Zenodo at the following DOI: https://doi.org/10.5281/zenodo.3991561.

Chianti is a precious red wine and enjoys a high reputation for its high quality in the world wine market. Despite this, the production region is small and product needs efficient tools to protect its brands and prevent adulterations. In this sense, ICP-MS combined with chemometrics has demonstrated its usefulness in food authentication. In this study, Chianti/Chianti Classico, authentic wines from vineyard of Toscana region (Italy), together samples from 18 different geographical regions, were analyzed with the objective of differentiate them from other Italian wines. Partial Least Squares-Discriminant Analysis (PLS-DA) identified variables to discriminate wine geographical origin. Rare Earth Elements (REE), major and trace elements all contributed to the discrimination of Chianti samples. General model was not suited to distinguish PDO red wines from samples, with similar chemical fingerprints, collected in some regions. Specific classification models enhanced the capability of discrimination, emphasizing the discriminant role of some elements.

Chianti is a precious red wine and enjoys a high reputation for its high quality in the world wine market. Despite this, the production region is small and product needs efficient tools to protect its brands and prevent adulterations. In this sense, ICP-MS combined with chemometrics has demonstrated its usefulness in food authentication. In this study, Chianti/Chianti Classico, authentic wines from vineyard of Toscana region (Italy), together samples from 18 different geographical regions, were analyzed with the objective of differentiate them from other Italian wines. Partial Least Squares-Discriminant Analysis (PLS-DA) identified variables to discriminate wine geographical origin. Rare Earth Elements (REE), major and trace elements all contributed to the discrimination of Chianti samples. General model was not suited to distinguish PDO red wines from samples, with similar chemical fingerprints, collected in some regions. Specific classification models enhanced the capability of discrimination, emphasizing the discriminant role of some elements.

Ranking and multi-criteria decision-making approaches are useful tools to analyse multivariate data and obtain useful insights into data structure and the relationships between samples and variables. In this study, we present a new ranking approach, named Deep Ranking Analysis by Power Eigenvectors (DRAPE), which is based on the Power-Weakness Ratio analysis and provides a set of sequential rankings. Such a sequential ranking procedure allows to gather deeper insights into the analysed dataset. Moreover, by a “retro”-regression procedure, the relevance of each variable in determining the final rankings can be assessed, while a consensus ranking can be obtained by a Principal Component Analysis (PCA). In this study, we present the theory of the novel method, and show three applications to real datasets.

The nuclear androgen receptor (AR) is one of the most relevant biological targets of Endocrine Disrupting Chemicals (EDCs), which produce adverse effects by interfering with hormonal regulation and endocrine system functioning. This paper describes novel in silico models to identify organic AR modulators in the context of the Collaborative Modeling Project of Androgen Receptor Activity (CoMPARA), coordinated by the National Center of Computational Toxicology (U.S. Environmental Protection Agency). The collaborative project involved 35 international research groups to prioritize the experimental tests of approximatively 40k compounds, based on the predictions provided by each participant. In this paper, we describe our machine learning approach to predict the binding to AR, which is based on a consensus of a multivariate Bernoulli Naive Bayes, a Random Forest, and N-Nearest Neighbor classification models. The approach was developed in compliance with the Organization of Economic Cooperation and Development (OECD) principles, trained on 1687 ToxCast molecules classified according to 11 in vitro assays, and further validated on a set of 3,882 external compounds. The models provided robust and reliable predictions and were used to gather novel data-driven insights on the structural features related to AR binding, agonism, and antagonism

This work presents the first-time QSAR approach to predict the laboratory-based fish biomagnification factor (BMF) of organic chemicals, to be used as a supporting tool for assessing bioaccumulation at the regulatory level. The developed strategy is based on 2 levels of prediction, with a varying trade-off between interpretability and performance according to the user's needs. We designed our models to be intrinsically acceptable at the regulatory level (in what we defined as “acceptable-by-design” strategy), by (i) complying with OECD principles directly in the approach development phase, (ii) choosing easy-to-apply modeling techniques, (iii) preferring simple descriptors when possible, and (iv) striving to provide data-driven mechanistic insights. Our novel tool has an error comparable to the observed experimental inter- and intraspecies variability and is stable on borderline compounds (root mean square error [RMSE] ranging from RMSE = 0.45 to RMSE = 0.45 log units on test data). Additionally, the models’ molecular descriptors are carefully described and interpreted, allowing us to gather additional mechanistic insights into the structural features controlling the dietary bioaccumulation of chemicals in fish. To improve the transparency and promote the application of the model, the data set and the stand alone prediction tool are provided free of charge at https://github.com/grisoniFr/bmf_qsar Integr Environ Assess Manag 2019;15:51–63. © 2018 SETAC.

The ICCVAM Acute Toxicity Workgroup (U.S. Department of Health and Human Services), in collaboration with the U.S. Environmental Protection Agency (U.S. EPA, National Center for Computational Toxicology), coordinated the “Predictive Models for Acute Oral Systemic Toxicity” collaborative project to develop in silico models to predict acute oral systemic toxicity for filling regulatory needs. In this framework, new Quantitative Structure-Activity Relationship (QSAR) models for the prediction of very toxic (LD50 lower than 50 mg/kg) and nontoxic (LD50 greater than or equal to 2,000 mg/kg) endpoints were developed, as described in this study. Models were developed on a large set of chemicals (8992), provided by the project coordinators, considering the five OCED principles for QSAR applicability to regulatory endpoints. A Bayesian consensus approach integrating three different classification QSAR algorithms was applied as modelling method. For both the considered endpoints, the proposed approach demonstrated to be robust and predictive, as determined by a blind validation on a set of external molecules provided in a later stage by the coordinators of the collaborative project. Finally, the integration of predictions obtained for the very toxic and nontoxic endpoints allowed the identification of compounds associated to medium toxicity, as well as the analysis of consistency between the predictions obtained for the two endpoints on the same molecules. Predictions of the proposed consensus approach will be integrated with those originated from models proposed by the participants of the collaborative project to facilitate the regulatory acceptance of in-silico predictions and thus reduce or replace experimental tests for acute toxicity.

Quantitative Structure – Activity Relationship (QSAR) models play a central role in medicinal chemistry, toxicology and computer-assisted molecular design, as well as a support for regulatory decisions and animal testing reduction. Thus, assessing their predictive ability becomes an essential step for any prospective application. Many metrics have been proposed to estimate the model predictive ability of QSARs, which have created confusion on how models should be evaluated and properly compared. Recently, we showed that the metric Q2F3 is particularly well-suited for comparing the external predictivity of different models developed on the same training dataset. However, when comparing models developed on different training data, this function becomes inadequate and only dispersion measures like the root-mean-square error (RMSE) should be used. The intent of this work is to provide clarity on the correct and incorrect uses of Q2F3, discussing its behavior towards the training data distribution and illustrating some cases in which Q2F3 estimates may be misleading. Hereby, we encourage the usage of measures of dispersions when models trained on different datasets have to be compared and evaluated

Legislators have included bioaccumulation in the evaluation of chemicals in the framework of the European Registration, Evaluation, Authorization, and Restriction of Chemicals (REACH) regulation. REACH requires information on the bioconcentration factor (BCF), which is a parameter for assessing bioaccumulation and encourages the use of a weight-of-evidence approach, including predictions from quantitative structure-activity relationships (QSARs). This study presents a novel approach, based on structural alerts, to be used as a decision-support system for the identification of substances with bioaccumulation potential. In a regulatory framework, these alerts can be integrated with other sources of information, such as experimental and in silico data, to reduce the uncertainty of the assessment, thereby supporting a weight-of-evidence approach. Moreover, the identified alerts have a direct connection with relevant structural features, thus fostering the applicability and interpretability of the approach. The structural alerts were identified on 779 chemicals annotated for their fish BCF, and the approach was then validated on 278 external molecules. The developed decision-support system allowed identification of 77% of bioaccumulative chemicals and was competitive with more complex QSAR models used in regulatory assessments. The approach is implemented in an easy-to-use workflow, provided free of charge. Integr Environ Assess Manag 2018;00:000-000. © 2018 SETAC.

Generative artificial intelligence offers a fresh view on molecular design. We present the first-time prospective application of a deep learning model for designing new druglike compounds with desired activities. For this purpose, we trained a recurrent neural network to capture the constitution of a large set of known bioactive compounds represented as SMILES strings. By transfer learning, this general model was fine-tuned on recognizing retinoid X and peroxisome proliferator-activated receptor agonists. We synthesized five top-ranking compounds designed by the generative model. Four of the compounds revealed nanomolar to low-micromolar receptor modulatory activity in cell-based assays. Apparently, the computational model intrinsically captured relevant chemical and biological knowledge without the need for explicit rules. The results of this study advocate generative artificial intelligence for prospective de novo molecular design, and demonstrate the potential of these methods for future medicinal chemistry

Natural products offer unexplored molecular frameworks for the development of chemical leads and innovative drugs. However, the structural complexity of natural products compared with synthetic drug-like molecules often limits the scaffold hopping potential of natural-product-inspired molecular design. Here we introduce a holistic molecular representation incorporating pharmacophore and shape patterns, which facilitates scaffold hopping from natural products to isofunctional synthetic compounds. This computational approach captures simultaneously the partial charge, atom distributions and molecular shape. In a prospective application, we use four natural cannabinoids as queries in a chemical database search for novel synthetic modulators of human cannabinoid receptors. Of the synthetic compounds selected by the new method, 35% are experimentally confirmed as active. These cannabinoid receptor modulators are structurally less complex than their respective natural product templates. The results of this study validate this holistic molecular representation for hit and lead finding in drug discovery

The characterization of 72 Italian honey samples from 8 botanical varieties was carried out by a comprehensive approach exploiting data fusion of IR, NIR and Raman spectroscopies, Proton Transfer Reaction – Time of Flight – Mass Spectrometry (PTR-MS) and electronic nose. High-, mid- and low-level data fusion approaches were tested to verify if the combination of several analytical sources can improve the classification ability of honeys from different botanical origins. Classification was performed on the fused data by Partial Least Squares – Discriminant Analysis; a strict validation protocol was used to estimate the predictive performances of the models. The best results were obtained with high-level data fusion combining Raman and NIR spectroscopy and PTR-MS, with classification performances better than those obtained on single analytical sources (accuracy of 99% and 100% on test and training samples respectively). The combination of just three analytical sources assures a limited time of analysis.

Molecular descriptors encode a wide variety of molecular information and have become the support of many contemporary chemoinformatic and bioinformatic applications. They grasp specific molecular features (e.g., geometry, shape, pharmacophores, or atomic properties) and directly affect computational models, in terms of outcome, performance, and applicability. This chapter aims to illustrate the impact of different molecular descriptors on the structural information captured and on the perceived chemical similarity among molecules. After introducing the fundamental concepts of molecular descriptor theory and application, a step-by-step retrospective virtual screening procedure guides users through the fundamental processing steps and discusses the impact of different types of molecular descriptors.

Molecular descriptors capture diverse parts of the structural information of molecules and they are the support of many contemporary computer-assisted toxicological and chemical applications. After briefly introducing some fundamental concepts of structure-activity applications (e.g., molecular descriptor dimensionality, classical vs. fingerprint description, and activity landscapes), this chapter guides the readers through a step-by-step explanation of molecular descriptors rationale and application. To this end, the chapter illustrates a case study of a recently published application of molecular descriptors for modeling the activity on cytochrome P450.

Lipophilicity prediction is routinely applied to small molecules and presents a working alternative to experimental log P or log D determination. For compounds outside the domain of classical medicinal chemistry these predictions lack accuracy, advocating the development of bespoke in silico approaches. Peptides and their derivatives and mimetics fill the structural gap between small synthetic drugs and genetically engineered macromolecules. Here, we present a data-driven machine learning method for peptide log D7.4 prediction. A model for estimating the lipophilicity of short linear peptides consisting of natural amino acids was developed. In a prospective test, we obtained accurate predictions for a set of newly synthesized linear tri- to hexapeptides. Further model development focused on more complex peptide mimetics from the AstraZeneca compound collection. The results obtained demonstrate the applicability of the new prediction model to peptides and peptide derivatives in a log D7.4 range of approximately −3 to 5, with superior accuracy to established lipophilicity models for small molecules

This work investigates the bioaccumulation patterns of 168 organic chemicals in fish, by comparing their bioconcentration factor (BCF), biomagnification factor (BMF) and octanol-water partitioning coefficient (KOW). It aims to gain insights on the relationships between dietary and non-dietary bioaccumulation in aquatic environment, on the effectiveness of KOW and BCF to detect compounds that bioaccumulate through diet, as well as to detect the presence of structure-related bioaccumulation patterns. A linear relationship between logBMF and logKOW was observed (logBMF = 1.14·logBCF – 6.20) up to logKOW ≈ 4, as well as between logBMF and logBCF (logBMF = 0.96·logBCF – 4.06) up to a logBCF ≈ 5. 10% of compounds do not satisfy the linear BCF-BMF relationship. The deviations from such linear relationships were further investigated with the aid of a self-organizing map and canonical correlation analysis, which allowed us to shed light on some structure-related patterns. Finally, the usage of KOW- and BCF-based thresholds to detect compounds that accumulate through diet led to many false positives (47%–91% for KOW), and a moderate number of false negatives (up to 5% for BCF). These results corroborate the need of using the experimental BMF for hazard assessment practices, as well as of developing computational tools for BMF prediction.

This special series is the outcome of the session "Predictive models in ecotoxicology: Bridging the gap between scientific progress and regulatory applicability," held at the 27th SETAC Europe annual meeting (Brussels, May 2017). In this foreword the rationale behind the special series, the reasons for proposing it, and its objectives are described briefly

The lack of potent subtype-selective modulators of retinoid X receptors (RXRs) has hindered their full exploitation as promising drug targets. Using computational similarity searching, target prediction and automated de novo design, we identified novel RXR ligands exhibiting innovative molecular frameworks, pronounced receptor-subtype preference and suitable properties for hit-to-lead expansion

A new chemoinformatic approach, called Mapping of Activity through Dichotomic Scores, is introduced. Its goal is the supervised projection of molecules, represented with strings of binary digits expressing the presence or absence of selected structural features, onto a novel 2-dimensional space, which highlights regions of active (inactive) molecules of interest. At the same time, variables are projected onto a second 2-dimensional space, which highlights those structural features that are more related to the molecular activity of interest. Unlike the classical weighting schemes used in substructural analysis, which consider the substructures independently of each other, the Mapping of Activity through Dichotomic Scores approach considers the interactions between pairs of substructures, that is, their frequencies of cooccurrence in the molecules. In this work, the theory is presented and elucidated, with an example dataset and in comparison with a benchmark fragment-based scoring scheme.

The assessment of the classification performance can be based on class indices, such as sensitivity, specificity and precision, which describe the classification results achieved on each modelled class. However, in several situations, it is useful to represent the global classification performance with a single number. Therefore, several measures have been introduced in literature to deal with this assessment, accuracy being the most known and used. These metrics have been proposed to generally face binary classification tasks and can behave differently depending on the classification scenario. In this study, different global measures of classification performances are compared by means of results achieved on an extended set of real multivariate datasets. The systematic comparison is carried out through multivariate analysis. Further investigations are then derived on specific indices to understand how the presence of unbalanced classes and the number of modelled classes can influence their behaviour. Finally, this work introduces a set of benchmark values based on different random classification scenarios. These benchmark thresholds can serve as the initial criterion to accept or reject a classification model on the basis of its performance

This work describes a novel approach based on advanced molecular similarity to predict the sweetness of chemicals. The proposed Quantitative Structure-Taste Relationship (QSTR) model is an expert system developed keeping in mind the five principles defined by the Organization for Economic Co-operation and Development (OECD) for the validation of (Q)SARs. The 649 sweet and non-sweet molecules were described by both conformation-independent extended-connectivity fingerprints (ECFPs) and molecular descriptors. In particular, the molecular similarity in the ECFPs space showed a clear association with molecular taste and it was exploited for model development. Molecules laying in the subspaces where the taste assignation was more difficult were modeled trough a consensus between linear and local approaches (Partial Least Squares-Discriminant Analysis and N-nearest-neighbor classifier). The expert system, which was thoroughly validated through a Monte Carlo procedure and an external set, gave satisfactory results in comparison with the state-of-the-art models. Moreover, the QSTR model can be leveraged into a greater understanding of the relationship between molecular structure and sweetness, and into the design of novel sweeteners.

Bioaccumulation is the process by which the concentration of a chemical in an organism exceeds that measured in the surrounding environment. It occurs through non‐dietary routes (e.g., through skin or respiratory surfaces) for what is known as bioconcentration, and/or through predator‐prey transport from diet. In most of national and international regulations, the detection of bioaccumulative chemicals relies on the bioconcentration capability. However, in some cases, the actual bioaccumulation can be more than 1 order of magnitude greater than that expected from the bioconcentration, due to storage through diet. This work targeted the development of a QSAR model to predict the bioaccumulation through the diet (i.e., Biomagnification Factor [BMF]). To allow for a regulatory application, the model was chosen as the best compromise between complexity and prediction accuracy, in terms of modelling method and interpretability of the molecular descriptors. The final model is a consensus of an Ordinary Least Squares and a k‐Nearest Neighbour models, which was developed on 214 compounds and led to accurate estimates of logBMF (RMSEPext = 0.45 l.u. and Q2ext = 0.82). To the best of our knowledge, no QSAR models for BMF developed on heterogeneous chemical sets currently exist. In addition, the work shows how the model and descriptor interpretations can be leveraged into a greater understanding of the structural features responsible for the underlying biological mechanisms

Molecular descriptors capture diverse structural information of molecules and are a prerequisite for ligand-based similarity searching. In this study, we introduce topological matrix-based descriptors to virtual screening for hit discovery. We evaluated the usefulness of matrix-based descriptors in a retrospective setting and compared them with topological pharmacophore descriptors. Special attention was given to the influence of data pre-processing and the applied similarity metric on the virtual screening performance. Overall, the MB descriptors showed a competitive and complementary performance to other descriptors. A prospective screen of a commercial compound library led to the discovery of a novel natural-product-derived cyclooxygenase-2 inhibitor predicted to interact differently with the target protein compared to the query compound ibuprofen. The results of our study motivate the use of matrix-based descriptors for molecular similarity-based virtual screening and scaffold hopping.

Similarity searching, a technique used for ligand-based virtual screening in drug discovery, exploits the structural information of known active compounds to find new ones with the desired biological activity. The effectiveness of similarity searching can be enhanced by combining independent searches, executed with different reference structures, through the socalled fusion rules. In this pilot study, we applied for the first time the weighted Power- Weakness Ratio (wPWR) as a fusion rule. wPWR was recently proposed as a multi-criteria decision-making and ranking comparison technique that, unlike existing parameter-free fusion rules, (1) is based on pairwise comparisons between compounds, and (2) can use weighting schemes. We compared wPWR with six fusion rules on four datasets using six evaluation metrics. The results indicated that wPWR has a mediocre performance but also the most robust behavior. Moreover, in one case, wPWR had the best accumulation curve on the top 25 compounds. These aspects are important for prospective applications. Weighting the reference structures according to their activity did not show a net beneficial effect on one analyzed dataset and, thus, further investigations are needed.

Cytochromes P450 (CYP) are the main actors in the oxidation of xenobiotics and play a crucial role in drug safety, persistence, bioactivation, and drug-drug/food-drug interaction. This work aims to develop Quantitative Structure-Activity Relationship (QSAR) models to predict the drug interaction with two of the most important CYP isoforms, namely 2C9 and 3A4. The presented models are calibrated on 9122 drug-like compounds, using three different modelling approaches and two types of molecular description (classical molecular descriptors and binary fingerprints). For each isoform, three classification models are presented, based on a different approach and with different advantages: (1) a very simple and interpretable classification tree; (2) a local (k-Nearest Neighbor) model based classical descriptors and; (3) a model based on a recently proposed local classifier (N-Nearest Neighbor) on binary fingerprints. The salient features of the work are (1) the thorough model validation and the applicability domain assessment; (2) the descriptor interpretation, which highlighted the crucial aspects of P450-drug interaction; and (3) the consensus aggregation of models, which largely increased the prediction accuracy.

This doctoral dissertation addresses the current open problems in the prediction of aquatic bioaccumulation through computational methods, which is a priority according to many national and international regulations. QSAR (Quantitative Structure-Activity Relationship) techniques are amongst the most-widely used in silico methods, as they (1) allow performing a quick analysis of hundreds of thousands of compounds, and (2) exploit advanced statistical and mathematical techniques for model development. QSAR is an effective tool for data-gap filling, testing prioritization and reduction of cost, time and number of animals used for experimental studies. The project addressed two issues (1) the non-dietary bioaccumulation of chemicals (expressed through the Bioconcentration Factor, BCF), which has been extensively modelled but still has some critical points, (2) the dietary bioaccumulation of chemicals, which was never modelled in a QSAR setting. Fish were used as the target, as they the benchmark organisms for bioaccumulation assessment. In the first phase, the performance of 9 BCF models of different complexity was analyzed and critically compared on a manually-curated dataset of fish BCFww for 1056 compounds. Results revealed that the descriptor-based models, despite their complexity, take only partially into account the processes that influence bioconcentration and show the same biases of KOW-based predictions (KOW = octanol-water partition coefficient). The KOW-driven over- and underestimation of BCF by are due to additional processes, such as metabolism or storage within non-lipid tissues. For this reason, a scheme was developed to predict whether a compound (1) is mainly stored within lipid tissues, (2) has additional storage sites (e.g., proteins), or (3) is metabolized/eliminated. The developed approach is based on two validated QSAR classification trees, characterized by descriptor interpretability and simplicity. They allowed to: (1) detect compounds that are p