Resumo em Inglês:

Natural Language Processing (NLP) in chemistry and materials science enables computers to understand, analyze, and generate human-readable output related to chemical concepts and materials. With the latest advancements in NLP, text processing at a near-human level has become possible in various tasks. Large Language Models (LLMs) have demonstrated exceptional proficiency in text generation, leading to the redefinition of numerous specific NLP tasks as text generation problems. This review explores the recent progress in applying LLMs to specialized domains, such as chemistry and materials science. It was discussed how LLMs overcome limitations of traditional NLP methods (such as rigid rule-based systems and shallow statistical models) by enabling context-aware interpretation of unstructured literature, flexible entity recognition (e.g., compounds, reactions), and generative tasks. Using the capabilities of LLMs, researchers in these fields can benefit from enhanced text processing, more accurate information extraction, and improved understanding of complex chemical concepts, making it a pivotal tool for accelerating discovery in chemically complex spaces, paving the way for novel tasks such as reaction prediction and molecular design.

Resumo em Inglês:

Theoretical and computational chemistry (TCC) is a set of theories and models that, over the years, were refined to the point that it is possible to determine measurable quantities with precision, predict experimental results, and provide fundamental insights into chemical phenomena and mechanisms that may be difficult or impossible to observe experimentally. Machine Learning (ML), on the other hand, is a subfield of Artificial Intelligence (AI) that applies different types of statistical methods to a large volume of data or a smaller volume of precise data, combined with high computational power, enabling the discovery of complex patterns and production of explanations inaccessible through human deductive reasoning and intuition alone or using traditional scientific methods. Recently, ML, combined or not with TCC methods, has emerged as a transformative force, bringing significant advances in chemistry and materials science. This review surveys basic ML concepts and their applications in chemistry, focusing on supervised and unsupervised learning approaches, data preprocessing, and model development workflows, exploring the most relevant ML algorithms selected for their specific usefulness in chemical applications. Integrating ML with traditional computational chemistry methods, such as density functional theory, is highlighted as a powerful synergy for accelerating materials discovery and design. Key areas of impact discussed include High-Throughput Virtual Screening (HTVS) of molecules and materials, spectroscopy (including UV-Vis and fluorescence), organic electronics (such as solar cells and organic light-emitting diodes), potential energy surfaces, and molecular dynamics. It also addresses critical aspects of ML in chemistry, including data representation, model interpretability through explainability techniques, and the emerging role of large language models (LLM). Tips for acquiring knowledge of ML for practical applications in chemistry are given.

Resumo em Inglês:

The integration of artificial neural networks (ANNs) with thermal analysis techniques, such as thermogravimetry (TG) and differential scanning calorimetry (DSC), presents great potential to accurately determine kinetics of thermal processes and thermal behavior. This review examines the application of ANNs, including Multilayer Perceptron (MLP) and Hopfield Neural networks (HNN), to analyze both isothermal and non-isothermal TG and DSC data. These networks provide robust tools for predicting thermal behavior, determining kinetic parameters, and modeling complex reaction mechanisms, significantly improving accuracy when compared to traditional methods. The review highlights advances in ANN-based methodologies, including the ability to integrate multiple kinetic models and manage noisy experimental data, thereby offering new insights into thermal decomposition and reaction kinetics. Moreover, the use of HNN in distributed activation energy models (DAEM) is explored, indicating their potential in understanding heterogeneous systems. This study emphasizes the transformative potential of ANNs for improving kinetic studies, suggesting future directions to explore machine learning in material characterization and thermal process optimization.

Resumo em Inglês:

Shikimate kinase emerges as an intriguing macromolecular target for the development of novel pharmaceutical agents for the treatment of tuberculosis. This study aimed to develop a neural network (NN) for the discovery of potential inhibitors of Mycobacterium tuberculosis shikimate kinase and to conduct molecular docking and molecular dynamics (MD) simulations. The NN model pointed out to a set of 810 molecules with anti-tuberculosis activity, wherein 86% of this set also demonstrated positive outcomes according to docking calculations. Among these, 54 molecules exhibited a docking score ranging from -9 to -9.8 kcal mol-1. Subsequently, a subset of molecules was selected for molecular dynamics studies and molecular mechanics Poisson-Boltzmann surface area (MM/PBSA) calculations. Furthermore, it was possible to assess that the dataset with higher affinity shared a similar electronic profile, as evidenced by the analysis of global descriptors (electronic chemical potential, hardness, and electrophilicity). The molecules displaying the lowest Gibbs free energy (∆G)binding values, therefore the highest affinity, were identified as CHEMBL1229147, CHEMBL4095667, and CHEMBL120640.

Resumo em Inglês:

Monkeypox (MPOX) is a zoonotic infectious disease caused by the monkeypox virus (MPXV) and has recently emerged as a significant concern for public health organizations globally. In 2022, the World Health Organization (WHO) reported thousands of laboratory confirmed cases, mobilizing the scientific community to control this phenomenon due to its emergency nature. Tecovirimat (TPOXX), a drug primarily recognized for the treatment of smallpox, has also been recommended for managing MPOX. It works by inhibiting the viral F13 protein (VP37), a critical component in the replication cycle of the virus. Some issues related to the possibility of drug resistance by the virus, the intrinsic chemical complexity of this molecule and the limited availability of therapeutic alternatives highlight the urgent need to explore and identify new effective compounds. In this paper, we propose the combination of modern machine learning techniques (deep reinforcement learning) with structure-based drug design (SBDD) approaches (molecular docking and dynamics) in the de novo design of molecular scaffolds with affinity for the F13 protein, lower structural complexity than TPOXX and easy synthetic accessibility, contributing to efforts in the search for therapeutic alternatives for MPOX.

Resumo em Inglês:

Leishmaniasis is a neglected tropical disease with limited therapeutic options characterized by high toxicity, adverse side effects, and growing resistance to existing treatments. In this study, machine learning (ML) methods were employed to design and evaluate benzophenone and xanthone derivatives as potential antileishmanial agents. A dataset of 73 compounds was curated, and Quantitative Structure-Activity Relationship (QSAR) models were developed using artificial neural networks (ANN), Random Forest (RF), and J48 decision tree classifiers. The ANN model achieved the highest accuracy (86.2%) in predicting antileishmanial activity, validated through in vitro assays. Among 14 newly synthesized benzophenones, compounds 5 and 7 demonstrated significant biological activity with inhibitory concentration 50 (IC50) values of 10.19 and 14.35 μM, respectively, and favorable selectivity indices compared to reference drugs pentamidine and amphotericin B. Structural analysis highlighted the importance of thiosemicarbazone and 4-methyl groups, alongside electronegative substituents at position 11, in enhancing activity. This study underscores the potential of computational tools to streamline the discovery of novel, effective, and selective antileishmanial agents.

Resumo em Inglês:

Chagas disease and African sleeping sickness are neglected tropical diseases (NTD) caused by Trypanosoma parasites, with current treatments facing challenges like toxicity and resistance. This study integrates machine learning and Quantitative Structure-Activity Relationship (QSAR) models to repurpose Food and Drug Administration (FDA)-approved drugs as potential treatments for these diseases. A dataset of 21,608 compounds with inhibitory activity against Trypanosoma cruzi and Trypanosoma brucei was analyzed using PubChem fingerprints. Random Forest and Extreme Gradient Boosting models were trained and applied to screen the ZINC-22 database for new therapeutic options. Posaconazole was predicted as the top candidate for multitarget activity against both Trypanosoma species, followed by pentamidine, a drug already approved for sleeping sickness. Additionally, 40 other drug candidates were identified by the models (pIC50 > 6 and coefficient of variation < 0.05), mainly antineoplastics (32%) and antifungals (19%). This approach demonstrates the potential of computational techniques in accelerating the discovery of drug candidates for neglected infectious diseases.

Resumo em Inglês:

The present study introduces a direct approach for classifying blood serum samples as either positive or negative for coronavirus disease (COVID-19) by associating the electrochemical impedance data of the sample with multivariate analysis. The hypothesis is that the systematic alterations in blood composition resulting from a severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection give rise to a distinct impedance spectrum when infected serum is subjected to analysis. A total of 201 serum samples were analyzed using the gold standard method, reverse transcription-polymerase chain reaction (RT-PCR), which served to train and validate the classification models. Two variations of discriminant analysis (partial least squares discriminant analysis (PLS-DA) and principal component analysis-discriminant analysis (PCA-DA)) and a one-class modeling approach (soft independent modeling of class analogies (SIMCA)) were used to classify impedance data in different formats (as complex or real numbers). PCA-DA applied to imaginary impedance spectra was found to be the most effective strategy, achieving sensitivity, specificity, and precision of 94, 94, and 91%, respectively, with classification error rates as low as 6%. These findings are encouraging and could facilitate the development of an inexpensive and reliable screening method for COVID-19.

Resumo em Inglês:

Accurate knowledge of electronic molecular properties of excited states is fundamental for understanding the behavior of functional materials for organic electronics and sensors. In this work, we focus on determining the properties of the most intense peak in the electronic absorption spectra of organic molecules. For this purpose, we employed the quantum chemistry QM-symex dataset, which has approximately 173,000 organic molecules and time-dependent density functional theory (TD-DFT) data of the first ten electronic absorption transitions. Each one is identified by its Cartesian coordinates. From data in the original QM-symex, we built a new dataset named QM symex-modif that contains molecules in simplified molecular input line entry system (SMILES) format and properties related to the main electronic transition. We then employed twenty machine learning (ML) algorithms to investigate oscillator strengths, excitation energies, transition orbitals, and the highest occupied molecular orbitals (HOMOs). As inputs for the ML algorithms, we used several chemical descriptors for each molecule generated in the RDKit tool employing the corresponding SMILES format. The generated input descriptors significantly improved the accuracy of the ML predictions for these key photophysical properties. Very good mean absolute errors (MAEs) were obtained for the test set composed of 45,056 molecules, namely, an MAE of 0.035 for oscillator strengths, 0.09 eV for excitation energies, 1.24 and 0.62 for the initial and final transition molecular orbital (MO) numbers (i.e., for each molecule, their position in the MO listing), respectively, and 0.014 for HOMO numbers, with coefficient of determination (R2) values consistently exceeding 0.94, thus demonstrating the accuracy of the models. Additionally, a Shapley additive explanation (SHAP) analysis was carried out to evaluate the importance of the input parameters for the investigated ML models. We found several interesting relationships involving the input parameters. In particular, molecular weight holds significant importance in our ML models for determining the target HOMO numbers and the transition orbitals.

Resumo em Inglês:

Label-free immunosensors based on screen-printed carbon electrodes offer a promising platform for the detection of cancer biomarkers. Herein, we explore the use of machine learning techniques to improve the performance of these immunosensors. We evaluate the influence of various redox probes on the analytical response in detecting the cancer biomarker protein p53. Ascorbic acid (AA) was found as the optimal redox probe, exhibiting a sensitivity of 0.26 ng mL-1, attributed to its strong affinity to proteins through hydrogen bonds and electrostatic interactions. We also extracted analytical information from the voltammograms, such as shifts in peak potential and changes in peak width, to construct datasets for supervised machine learning. Using different algorithms including logistic regression, linear discriminant analysis, K-nearest neighbor, Gaussian Naive-Bayes, decision trees, and support vector machine, we identified positive samples spiked with p53 in artificial urine and saliva samples. Through a comparison of immunosensors with distinct molecular architectures, we determined the critical role of redox probe selection, which proves to be more significant than modifying the working electrodes in determining performance. Furthermore, immunosensors with inferior inherent detection ability can achieve comparable performance to those with superior analytical characteristics when feature selection and machine learning algorithms are applied to the voltammograms. These findings illustrate the significance of extracting additional information from differential pulse voltammograms beyond peak current intensity. Furthermore, using machine learning techniques allows one to design biosensors capable of distinguishing biomarkers even in complex samples.

Resumo em Inglês:

This study investigates the emissions of biogenic volatile organic compounds (BVOCs) in an Atlantic Forest fragment and examines their correlation with abiotic factors. Despite its rich biodiversity, the Atlantic Forest remains under threat and lacks extensive research on BVOC emissions. Using supervised learning methods (including traditional Partial Least Squares Discriminant Analysis (PLS-DA), PLS-DA with bootstrap resampling, and Support Vector Machine Ensemble (SVM ensemble) 10 BVOCs were analyzed and correlated with environmental variables. The results reveal emission patterns linked to abiotic conditions, with the models achieving high classification accuracy. The variables a-pinene, linalool, and isobornyl acetate contribute more significantly to evening/morning samples, whereas temperature and wind gusts contribute more to afternoon samples. The PLS-DA and SVM ensemble models demonstrated effective sample classification, with minimal misclassification errors. The PLS-DA bootstrap model enhanced the robustness and reliability of sample classifications, confirming the consistent and distinct differences between sample classes. To the best of our knowledge, this is the first study to apply the SVM ensemble method to BVOCs in air samples and their correlation with abiotic factors, highlighting its potential to enhance the understanding of atmospheric dynamics and inform conservation strategies for the Caparaó region and other similarly threatened biomes.

Resumo em Inglês:

Active learning (AL) has been widely applied in chemistry and materials science. In this work, we propose a quantum active learning (QAL) method for automatic structural determination of doped nanoparticles, where quantum machine learning (QML) models for regression are used iteratively to indicate new structures to be calculated by Density Functional Theory (DFT) or Density Functional Based Tight Binding (DFTB) and this new data acquisition is used to retrain the QML models. The QAL method is implemented in the Quantum Machine Learning Software/Agent for Material Design and Discovery (QMLMaterial), whose aim is using an artificial agent (defined by QML regression algorithms) that chooses the next doped configuration to be calculated that has a higher probability of finding the optimum structure. The QAL uses a quantum Gaussian process with a fidelity quantum kernel as well as the projected quantum kernel and different quantum circuits. For comparison, classical AL was used with classical Gaussian process with different classical kernels. The presented QAL method was applied in the structural determination of doped Si11 with 4 Al (4Al@Si11) and the results indicate the QAL method is able to find the optimum 4Al@Si11 structure. The aim of this work is to present the QAL method, formulated in a noise-free quantum computing framework, for automatic structural determination of doped nanoparticles and materials defects.

Resumo em Inglês:

The development of predictive quantitative structure-activity relationship (QSAR) models using machine learning (ML) algorithms has become increasingly feasible due to the growing availability of chemical libraries with experimental data. These models can accelerate the drug discovery process and reduce failure rates by enabling data-driven decision-making. However, existing standalone software often lacks several critical components necessary for effective data preparation and modeling. Here, we introduce QSAR-Lit, an innovative, no-code, and comprehensive workflow designed for curating chemical and biological data, generating QSAR models, and performing virtual screening through an interactive Python-based Streamlit dashboard. The QSAR model development process begins with data curation, collecting and cleaning data on chemical structures and their biological activities. The next step is model building, where the curated data is used to train and optimize QSAR models. Finally, QSAR-Lit provides virtual screening, allowing QSAR models to predict the activity of new chemical structures. This application efficiently screens libraries of chemical compounds, assisting researchers in identifying and prioritizing potential candidates for further investigation.

Resumo em Inglês:

Applying machine learning (ML) models to predict the biological activity of molecules is a crucial approach to accelerate the development of new drugs, particularly for treating parasitic diseases such as schistosomiasis. This study explored various ML algorithms, including random forest (RF), support vector regression (SVR), extreme gradient boosting (XGBoost), and deep neural networks (DNN), to identify the most effective model for predicting the biological activity of potential bioactive compounds. RF emerged as the top-performing model, achieving a coefficient of determination (R2) of 0.93 in internal validation and 0.89 in external validation, while DNN achieved R2 values of 0.83 and 0.78, respectively. Feature importance analysis revealed that specific molecular fingerprints, such as those associated with fused aromatic systems (indoles and quinolines) and electron-donating groups (hydroxyl and methoxy), positively correlated with antiparasitic activity. In contrast, features linked to molecular flexibility negatively impacted activity. Using the RF model, 14 metabolites were selected from natural product databases, prioritizing those with high-predicted activity. These compounds belong to different classes of metabolites, predominantly alkaloids and flavonoids, which have presented efficacy against Schistosoma mansoni in the literature. The robust predictive models obtained in this study characterize the chemical properties of active compounds, offering valuable insights to enhance the drug discovery process and combat schistosomiasis.

Resumo em Inglês:

The limited treatment options for Chagas disease (CD), coupled with the side effects of approved drugs, highlight the urgent need to identify novel therapeutic candidates for this neglected disease. To address this challenge, we developed a strategy to assess whether existing data on this topic could support a comprehensive approach for predicting the antitrypanosomal properties of natural and synthetic compounds against Trypanosoma cruzi. The chemical space constructed in this study comprised various classes of compounds, which were subjected to machine learning-based screening to evaluate the robustness of different methodologies for handling heterogeneous data. The optimized models accurately predicted the antitrypanosomal activity of multiple compounds in the external test set, achieving correlation coefficient of the training set (R2) = 0.995, robustness (Q2) = 0.935, root mean squared error (RMSE) = 0.366, percentage split (P2) = 980, and correlation with predictions on the external test set (P2(test set)) = 0.888. Additionally, twenty-five compounds were tested in vitro against T. cruzi trypomastigotes, and the model successfully predicted their logarithm of half maximal inhibitory concentration (pIC50) values. This study proposes an approach for developing in house methodologies to identify promising candidates for biological assays by leveraging machine learning within a heterogeneous chemical space.

Resumo em Inglês:

The electrooxidation of glycerol offers a promising pathway for energy transition and biomass valorization, making it a key area of research. This study employs machine learning (ML) to predict the onset of glycerol electrooxidation and anodic peak potentials, enhancing the understanding of factors influencing these metrics. A dataset derived from 155 research articles includes parameters such as pH, electrolyte type, reference electrode, electrode material, current density, and scan rate. Fourteen ML algorithms were evaluated, with adaptive boosting (AdaBoost) achieving the best performance: root mean square error (RMSE) of 0.117 and coefficient of determination (R2) of 0.902 for onset potential and RMSE of 0.122 and R2 of 0.870 for anodic peak potential. Explainable artificial intelligence (XAI) techniques like Shapley additive explanations (SHAP) analysis identified pH, electrolyte type, and electrode properties (e.g., atomic number, electronegativity) as key predictors. Replacing elemental features with atomic properties improved performance and reduced complexity. This work demonstrates the potential of ML to optimize glycerol oxidation and advance alcohol electrooxidation research.

Resumo em Inglês:

The Ant Colony Optimization (ACO) algorithm is a nature-inspired metaheuristic method used for optimization problems. Although not a machine learning method per se, ACO is often employed alongside machine learning models to enhance performance through optimization. An ACO algorithm was adopted to optimize the so-called FVC density functional for the ground-state energy of strongly correlated systems. The parameter configurations that maximize optimization efficiency were found, while reducing the mean relative error (MRE) of the ACO functional. Then, it was analyzed the performance of the algorithm across different dimensionalities (1D-5D), which are related to the number of parameters to be optimized within the FVC functional. Our results indicate that 15 ants with a pheromone evaporation rate superior to 0.2 are sufficient to minimize the MRE for a vast regime of parameters of the strongly-correlated system - interaction, particle density, and spin magnetization. While the 1D, 2D, and 4D optimizations yield 1.5% < MRE < 2.7%, the 3D and 5D optimizations lower the MRE to ca. 0.8%, reflecting a 67% error reduction compared to the original FVC functional (MRE = 2.4%). As the simulation time grows almost linearly with dimension, our results highlight the potential of ant colony algorithms for density-functional problems, combining effectiveness with low computational cost.

Resumo em Inglês:

In the present work, three series of cinnamic (1), p-coumaric (2) and ferulic (3) esters containing different side-chains such as ethyl (1a-3a), n-propyl (1b-3b), n-butyl (1c-3c), n-pentyl (1d-3d), n-hexyl (1e-3e), and n-heptyl (1f-3f) were prepared, tested for activity against trypomastigote forms of the parasite Trypanosoma cruzi and toxicity against NCTC cells. Obtained results indicated that the presence of p-coumaric or ferulic moieties associated with C4-C7 linear side-chains play an important role in the bioactivity against T. cruzi since compounds 2c-2f and 3d-3f were found to be the most active derivatives with a half maximal effective concentration (EC50) value ranging from 12.8 to 1.7 μM, superior to that determined for the positive control benznidazole (EC50 = 16.4 μM). Additionally, machine learning and multivariate statistical analyses identified molecular features correlated with biological activity, emphasizing the importance of side-chain length and lipophilicity, highlighting the significance of the molecular structure of phenylpropanoid derivatives in the activity against T. cruzi.