Molecular Descriptors Research Articles

Predictions of Colloidal Molecular Aggregation Using AI/ML Models.

To facilitate the triage of hits from small molecule screens, we have used various AI/ML techniques and experimentally observed data sets to build models aimed at predicting colloidal aggregation of small organic molecules in aqueous solution. We have found that Naïve Bayesian and deep neural networks outperform logistic regression, recursive partitioning tree, support vector machine, and random forest techniques by having the lowest balanced error rate (BER) for the test set. Derived predictive classification models consistently and successfully discriminated aggregator molecules from nonaggregator hits. An analysis of molecular descriptors in favor of colloidal aggregation confirms previous observations (hydrophobicity, molecular weight, and solubility) in addition to undescribed molecular descriptors such as the fraction of sp3 carbon atoms (Fsp3), and electrotopological state of hydroxyl groups (ES_Sum_sOH). Naïve Bayesian modeling and scaffold tree analysis have revealed chemical features/scaffolds contributing the most to colloidal aggregation and nonaggregation, respectively. These results highlight the importance of scaffolds with high Fsp3 values in promoting nonaggregation. Matched molecular pair analysis (MMPA) has also deciphered context-dependent substitutions, which can be used to design nonaggregator molecules. We found that most matched molecular pairs have a neutral effect on aggregation propensity. We have prospectively applied our predictive models to assist in chemical library triage for optimal plate selection diversity and purchase for high throughput screening (HTS) in drug discovery projects.

Machine learning models and performance dependency on 2D chemical descriptor space for retention time prediction of pharmaceuticals

The predictive modeling of liquid chromatography methods can be an invaluable asset, potentially saving countless hours of labor while also reducing solvent consumption and waste. Tasks such as physicochemical screening and preliminary method screening systems where large amounts of chromatography data are collected from fast and routine operations are particularly well suited for both leveraging large datasets and benefiting from predictive models. Therefore, the generation of predictive models for retention time is an active area of development. However, for these predictive models to gain acceptance, researchers first must have confidence in model performance and the computational cost of building them should be minimal. In this study, a simple and cost-effective workflow for the development of machine learning models to predict retention time using only Molecular Operating Environment 2D descriptors as input for support vector regression is developed. Furthermore, we investigated the relative performance of models based on molecular descriptor space by utilizing uniform manifold approximation and projection and clustering with Gaussian mixture models to identify chemically distinct clusters. Results outlined herein demonstrate that local models trained on clusters in chemical space perform equivalently when compared to models trained on all data. Through 10-fold cross-validation on a comprehensive set containing 67,950 of our company's proprietary analytes, these models achieved coefficients of determination of 0.84 and 3 % error in terms of retention time. This promising statistical significance is found to translate from cross-validation to prospective prediction on an external test set of pharmaceutically relevant analytes. The observed equivalency of global and local modeling of large datasets is retained with METLIN's SMRT dataset, thereby confirming the wider applicability of the developed machine learning workflows for global models.

Molecular temperature descriptors as a novel approach for QSPR analysis of Borophene nanosheets.

Borophene nanosheets appear in various sizes and shapes, ranging from simple planar structures to complicated polyhedral formations. Due to their unique chemical, optical, and electrical properties, Borophene nanosheets are theoretically and practically attractive and because of their high thermal conductivity, boron nanosheets are suitable for efficient heat transmission applications. In this paper, temperature indices of borophene nanosheets are computed and these indices are employed in QSPR analysis of attributes like Young's modulus, Shear modulus, and Poisson's ratio of borophene nanosheets and borophene β12 sheets. The regression model for the F-Temperature index is discovered to be the best fit for shear modulus, the reciprocal product connectivity temperature index is discovered to be fit for Poisson's ratio and the second hyper temperature index is discovered to be fit for Young's modulus based on the correlation coefficient.

Extrapolative Machine Learning for Accurate Efficiency Prediction in Non‐Fullerene Ternary Organic Solar Cells: Leveraging Computable Molecular Descriptors in High‐Throughput Virtual Screening

Adding a third component to binary organic solar cells (OSCs) enhances ternary OSCs, boosting power conversion efficiency (PCE). However, developing and optimizing appropriate donors, acceptors, and ternary materials remains a complex and demanding task. This study presents four machine‐learning (ML) predictive models using XGBoost and ANN approaches, utilizing both experimental and DFT‐derived HOMO and LUMO levels for efficient high‐throughput virtual screening (HTVS) of top candidates based on PCE. Two distinct latent databases were employed for HTVS: one consisting of 429 413 uniquely recombined ternary OSC systems from experimentally available data, and another comprising ≈2.3 million unique donor molecules from the Harvard Clean Energy Project database (CEPDB). The four ML models demonstrated notable predictive accuracy for PCE on a test dataset containing molecules closely aligned with the training set (interpolation). However, the XGBoost model showed constrained extrapolative ability for molecules significantly divergent from those in the training dataset. In contrast, the ANN models displayed a robust extrapolative capacity in HTVS, successfully predicting new potential ternary OSC systems and leading donors with PCE values exceeding 20%. Our ML models use HOMO and LUMO inputs for donors, acceptors, and ternaries, facilitating efficient optimization via rapid HTVS of high‐performance ternary materials.

Understanding the phytotoxic effects of organic contaminants on rice through predictive modeling with molecular descriptors: A data-driven analysis

The widespread introduction of organic compounds into environments poses significant risks to ecosystems. Assessing the adverse effects of organic contaminants on crops is crucial for ensuring food safety. However, laboratory research is often time-consuming and costly, and machine learning (ML) methods can offer a viable solution to address these challenges. This study aimed at developing a ML model that incorporates chemical descriptors to predict the phytotoxicity of organic contaminants on rice. A dataset was compiled by gathering published experimental data on the phytotoxicity of 60 organic compounds, with a focus on morphological inhibition, photosynthesis perturbation, and oxidative stress. Four ML models (RF, SVM, GBM, ANN) were developed using chemical molecular descriptors (CMD) and the Molecular ACCess System (MACCS) keys. RF-MACCS model demonstrated the highest fitness, achieving an R2 value of 0.79 and an RMSE of 0.14. Feature importance analysis highlighted nAtom, HBA, logKow, and TPSA as the most influential CMDs in our model. Additionally, substructures containing oxygen atoms, carbonyl group and carbon chains with nitrogen and oxygen atoms were identified as significant factors associated with phytotoxicity. This data-driven study could aid in predicting the phytotoxicity of organic contaminants on crops and evaluating the potential risks of emerging contaminants in agroecosystems.

From molecular descriptors to the developmental toxicity prediction of pesticides/veterinary drugs/bio-pesticides against zebrafish embryo: Dual computational toxicological approaches for prioritization

The escalating introduction of pesticides/veterinary drugs into the environment has necessitated a rapid evaluation of their potential risks to ecosystems and human health. The developmental toxicity of pesticides/veterinary drugs was less explored, and much less the large-scale predictions for untested pesticides, veterinary drugs and bio-pesticides. Alternative methods like quantitative structure-activity relationship (QSAR) are promising because their potential to ensure the sustainable and safe use of these chemicals. We collected 133 pesticides and veterinary drugs with half-maximal active concentration (AC50) as the zebrafish embryo developmental toxicity endpoint. The QSAR model development adhered to rigorous OECD principles, ensuring that the model possessed good internal robustness (R2 > 0.6 and QLOO2 > 0.6) and external predictivity (Rtest2 > 0.7, QFn2 >0.7, and CCCtest > 0.85). To further enhance the predictive performance of the model, a quantitative read-across structure-activity relationship (q-RASAR) model was established using the combined set of RASAR and 2D descriptors. Mechanistic interpretation revealed that dipole moment, the presence of C-O fragment at 10 topological distance, molecular size, lipophilicity, and Euclidean distance (ED)-based RA function were main factors influencing toxicity. For the first time, the established QSAR and q-RASAR models were combined to prioritize the developmental toxicity of a vast array of true external compounds (pesticides/veterinary drugs/bio-pesticides) lacking experimental values. The prediction reliability of each query molecule was evaluated by leverage approach and prediction reliability indicator. Overall, the dual computational toxicology models can inform decision-making and guide the design of new pesticides/veterinary drugs with improved safety profiles.

Effect of the central atom and donor on linear and nonlinear optical properties of tricyanopyrrolidone derivatives: DFT and TD-DFT approach

We have designed 50 molecules having “donor-π-acceptor system.” Different donors connected to tricyanopyrrolidone acceptor with 10 different heteroatom units are selected to study their “ground state” structural and nonlinear optical (NLO) properties. Frontier molecular orbital (FMO), molecular electrostatic potential (MEP) plots, and bond length alteration (BLA) showed intramolecular charge transfer (ICT). Results suggest that HOMO-LUMO energies of selected dyes depend on donor strength and heteroatoms. Triphenylamine substituted donors showed red-shifted vertical excitations, higher dipole moments, and higher hyperpolarizabilities. Using DFT, NLO parameters such as (β0) and (γ) are calculated with three different functionals, including global hybrid (GH), B3LYP and range separated CAM-B3LYP, and ωB97XD functionals. Molecular reactivity descriptors, chemical potential (μ), chemical hardness (η), electrophilicity index (ω), and hyperhardness (Γ) are produced. DFT descriptors are correlated with calculated NLO parameters.

SB-Net: Synergizing CNN and LSTM networks for uncovering retrosynthetic pathways in organic synthesis

Retrosynthesis is vital in synthesizing target products, guiding reaction pathway design crucial for drug and material discovery. Current models often neglect multi-scale feature extraction, limiting efficacy in leveraging molecular descriptors. Our proposed SB-Net model, a deep-learning architecture tailored for retrosynthesis prediction, addresses this gap. SB-Net combines CNN and Bi-LSTM architectures, excelling in capturing multi-scale molecular features. It integrates parallel branches for processing one-hot encoded descriptors and ECFP, merging through dense layers. Experimental results demonstrate SB-Net’s superiority, achieving 73.6 % top-1 and 94.6 % top-10 accuracy on USPTO-50k data. Versatility is validated on MetaNetX, with rates of 52.8 % top-1, 74.3 % top-3, 79.8 % top-5, and 83.5 % top-10. SB-Net’s success in bioretrosynthesis prediction tasks indicates its efficacy. This research advances computational chemistry, offering a robust deep-learning model for retrosynthesis prediction. With implications for drug discovery and synthesis planning, SB-Net promises innovative and efficient pathways.

Machine learning-enabled discovery of multi-resonance TADF molecules: Unraveling PLQY predictions from molecular structures

Unlocking the potential of multi-resonance thermally activated delayed fluorescence (MR-TADF) molecules for advanced organic light-emitting diode applications requires an insightful understanding of the relationship between molecular structures and photoluminescence quantum yield (PLQY). Utilizing molecular descriptors as inputs for machine learning (ML) algorithms, further illuminated by SHapley Additive exPlanations (SHAP) to interpret the ML model outcomes, this method effectively connects molecular structures to PLQY, providing targeted guidance for molecular design. A vast molecular library is generated via variational autoencoders, allowing for a comprehensive exploration of molecular space beyond conventional chemical intuition. High-throughput virtual screening, combined with our PLQY-focused model and a secondary model for emission peak wavelength prediction, efficiently identify promising candidates with blue-emitting properties. The robustness of our predictions is substantiated through quantum chemistry calculations. The integrative methodology proposed in this work not only streamlines the discovery of MR-TADF molecules but also provides a replicable framework for the intelligent design of other optoelectronic materials.

Machine Learned Classification of Ligand Intrinsic Activities at Human μ-Opioid Receptor.

Opioids are small-molecule agonists of μ-opioid receptor (μOR), while reversal agents such as naloxone are antagonists of μOR. Here we developed machine learning (ML) models to classify the intrinsic activities of ligands at the human μOR based on the SMILE strings and two-dimensional molecular descriptors. We first manually curated a database of 983 small molecules with measured E max values at the human μOR. Analysis of the chemical space allowed identification of dominant scaffolds and structurally similar agonists and antagonists. Decision tree models and directed message passing neural networks (MPNNs) were then trained to classify agonistic and antagonistic ligands. The hold-out test AUCs (areas under the receiver operator curves) of the extra-tree (ET) and MPNN models are 91.5±3.9% and 91.8± 4.4%, respectively. To overcome the challenge of small dataset, a student-teacher learning method called tri-training with disagreement was tested using an unlabeled dataset comprised of 15,816 ligands of human, mouse, or rat μOR, κOR, or δOR. We found that the tri-training scheme was able to increase the hold-out AUC of MPNN to as high as 95.7%. Our work demonstrates the feasibility of developing ML models to accurately predict the intrinsic activities of μOR ligands, even with limited data. We envisage potential applications of these models in evaluating uncharacterized substances for public safety risks and discovering new therapeutic agents to counteract opioid overdoses.

Predictions of biorelevant solubility change during dispersion and digestion of lipid-based formulations

Computational approaches are increasingly explored in development of drug products, including the development of lipid-based formulations (LBFs), to assess their feasibility for achieving adequate oral absorption at an early stage. This study investigated the use of computational pharmaceutics approaches to predict solubility changes of poorly soluble drugs during dispersion and digestion in biorelevant media. Concentrations of 30 poorly water-soluble drugs were determined pre- and post-digestion with in-line UV probes using the MicroDISS Profiler™. Generally, cationic drugs displayed higher drug concentrations post-digestion, whereas for non-ionized drugs there was no discernible trend between drug concentration in dispersed and digested phase. In the case of anionic drugs there tended to be a decrease or no change in the drug concentration post-digestion. Partial least squares modelling was used to identify the molecular descriptors and drug properties which predict changes in solubility ratio in long-chain LBF pre-digestion (R2 of calibration = 0.80, Q2 of validation = 0.64) and post-digestion (R2 of calibration = 0.76, Q2 of validation = 0.72). Furthermore, multiple linear regression equations were developed to facilitate prediction of the solubility ratio pre- and post-digestion. Applying three molecular descriptors (melting point, LogD, and number of aromatic rings) these equations showed good predictivity (pre-digestion R2 = 0.70, and post-digestion R2 = 0.68). The model developed will support a computationally guided LBF strategy for emerging poorly water-soluble drugs by predicting biorelevant solubility changes during dispersion and digestion. This facilitates a more data-informed developability decision making and subsequently facilitates a more efficient use of formulation screening resources.

Descriptors-based machine-learning prediction of cetane number using quantitative structure–property relationship

The physicochemical properties of liquid alternative fuels are important but difficult to measure/predict, especially when complex surrogate fuels are concerned. In the present work, machine learning is used to develop quantitative structure–property relationship models. The fuel chemical structure is represented by molecular descriptors, allowing the linking of important features of the fuel composition and key properties of fuel utilization. Feature selection is employed to select the most relevant features that describe the chemical structure of the fuel and several machine learning algorithms are tested to construct interpretable models. The effectiveness of the methodology is demonstrated through the development of accurate and interpretable predictive models for cetane numbers, with a focus on understanding the link between molecular structure and fuel properties. In this context, matrix-based descriptors and descriptors related to the number of atoms in the molecule are directly linked with the cetane number of hydrocarbons. Furthermore, the results showed that molecular connectivity indices play a role in the cetane number for aromatic molecules. Also, the methodology is extended to predict the cetane number of ester and ether molecules, leveraging the design of alternative fuels towards fully sustainable fuel utilization.

DeepRA: A novel deep learning-read-across framework and its application in non-sugar sweeteners mutagenicity prediction

Non-sugar sweeteners (NSSs) or artificial sweeteners have long been used as food chemicals since World War II. NSSs, however, also raise a concern about their mutagenicity. Evaluating the mutagenic ability of NSSs is crucial for food safety; this step is needed for every new chemical registration in the food and pharmaceutical industries. A computational assessment provides less time, money, and involved animals than the in vivo experiments; thus, this study developed a novel computational method from an ensemble convolutional deep neural network and read-across algorithms, called DeepRA, to classify the mutagenicity of chemicals. The mutagenicity data were obtained from the curated Ames test data set. The DeepRA model was developed using both molecular descriptors and molecular fingerprints. The obtained DeepRA model provides accurate and reliable mutagenicity classification through an independent test set. This model was then used to examine the NSSs-related chemicals, enabling the evaluation of mutagenicity from the NSSs-like substances. Finally, this model was publicly available at https://github.com/taraponglab/deepra for further use in chemical regulation and risk assessment.

New QSPR molecular descriptors based on low-cost quantum chemistry computations using DFT/COSMO approach

This work presents a simple computational methodology to determine new theoretical molecular descriptor scales convenient for use in QSPR correlations and the analysis of solvation-related properties. On the basis of low-cost quantum chemical computations using DFT/COSMO approach, in particular that implemented in the ADF/COSMO-RS module of the Amsterdam Modeling Studio program package, optimized geometry and local screening charge density of a considered molecule are obtained and related to the proposed descriptors through an a priori simple reasoning. Values of four molecular descriptors, namely volume VCOSMO∗, hydrogen bond/Lewis acidity αCOSMO and basicity βCOSMO, and charge asymmetry of the nonpolar region of the molecule δCOSMO have been calculated and are presented for sets of 128 non-ionic organic molecules and 47 ions composing ionic liquids. Relation of the proposed scales to some others that are widely known or presented recently in the literature has been examined. Although the present methodology is completely independent of any experimental data, the theoretical descriptor scales proposed here have been found to correlate linearly quite well with the respective empirical scales (mostly R2 > 0.8, and for some R2 > 0.9). For non-ionic organics, just the direct proportionality provided adequate fits of the well-established prominent scales such as those of Abraham, Klamt, Kamlet-Taft, Catalan, Gutmann, and Laurence and showed only a few statistically justifiable outliers. For ions composing ILs, the linear fits of seven acidity or basicity scales presented previously with those proposed in this work were of similar quality. Examining the observed outliers, several descriptor values reported in the literature were identified as being in error. Good performance of the new descriptor scales has been demonstrated by employing them for LSER fitting of various solvation-related thermodynamic and kinetic properties such as standard vaporization enthalpy, standard hydration enthalpy, air–water partition coefficient, air-IL partition coefficient, and solvent effects on the activation Gibbs energy or rate constant of SN1 and SNAr reactions. The quality of these correlations appears to be comparable to that of the correlations currently available for these properties, regardless of the role, solute or solvent, the involved molecules played.

Sequential closed-loop Bayesian optimization as a guide for organic molecular metallophotocatalyst formulation discovery

Conjugated organic photoredox catalysts (OPCs) can promote a wide range of chemical transformations. It is challenging to predict the catalytic activities of OPCs from first principles, either by expert knowledge or by using a priori calculations, as catalyst activity depends on a complex range of interrelated properties. Organic photocatalysts and other catalyst systems have often been discovered by a mixture of design and trial and error. Here we report a two-step data-driven approach to the targeted synthesis of OPCs and the subsequent reaction optimization for metallophotocatalysis, demonstrated for decarboxylative sp3–sp2 cross-coupling of amino acids with aryl halides. Our approach uses a Bayesian optimization strategy coupled with encoding of key physical properties using molecular descriptors to identify promising OPCs from a virtual library of 560 candidate molecules. This led to OPC formulations that are competitive with iridium catalysts by exploring just 2.4% of the available catalyst formulation space (107 of 4,500 possible reaction conditions).

In Vivo Ocular Pharmacokinetics and Toxicity of Siponimod in Albino Rabbits.

Siponimod is a promising agent for the inhibition of ocular neovascularization in diabetic retinopathy and age-related macular degeneration. Siponimod's development for ophthalmological application is hindered by the limited information available on the drug's solubility, stability, ocular pharmacokinetics (PK), and toxicity in vivo. In this study, we investigated the aqueous stability of siponimod under stress conditions (up to 60 °C) and its degradation behavior in solution. Additionally, siponimod's ocular PK and toxicity were investigated using intravitreal injection of two different doses (either 1300 or 6500 ng) in an albino rabbit model. Siponimod concentration was quantified in the extracted vitreous, and the PK parameters were calculated. The drug half-life after administration of the low and high doses was 2.8 and 3.9 h, respectively. The data obtained in vivo was used to test the ability of published in silico models to predict siponimod's PK accurately. Two models that correlated siponimod's molecular descriptors with its elimination from the vitreous closely predicted the half-life. Furthermore, 24 h and 7 days after intravitreal injections, the retinas showed no signs of toxicity. This study provides important information necessary for the formulation and development of siponimod for ophthalmologic applications. The short half-life of siponimod necessitates the development of a sustained drug delivery system to maintain therapeutic concentrations over an extended period, while the lack of short-term ocular toxicity observed in the retinas of siponimod-treated rabbits supports possible clinical use.

Breast Cancer Drugs Screening Model Based on Graph Convolutional Network and Ensemble Method

Breast cancer is the first cancer incidence and the second cancer mortality in women. Therefore, for the life and health of breast cancer patients, the research and development of breast cancer drugs should be accelerated. In drug development, the search for compounds with good bioactivity, pharmacokinetics, and safety, including Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET), has always been a time-consuming and labor-intensive process. In this paper, the relationship between the molecular descriptor and ADMET properties of compounds is studied. Aiming at the problem of composite ADMET attribute classification, a Stacking Algorithm based on Graph Convolutional Network (SA-GCN) was proposed. Firstly, feature selection was performed in the data of molecular descriptors. Then the SA-GCN is developed by integrating the advantages of ten classical classification algorithms. Finally, various performance indicators were used to conduct comparative experiments. Experiments show that the SA-GCN is superior to other classifiers in the classification performance of ADMET, and the classification accuracy is 97.6391%, 98.1450%, 94.4351%, 96.4587%, and 97.9764% compared to other classifiers. Therefore, this method can be well applied to the classification of ADMET properties of compounds and then could provide some help to screen out compounds with good biological activities.

A Machine Learning Approach for Predicting Caco-2 Cell Permeability in Natural Products from the Biodiversity in Peru.

Peru is one of the most biodiverse countries in the world, which is reflected in its wealth of knowledge about medicinal plants. However, there is a lack of information regarding intestinal absorption and the permeability of natural products. The human colon adenocarcinoma cell line (Caco-2) is an in vitro assay used to measure apparent permeability. This study aims to develop a quantitative structure-property relationship (QSPR) model using machine learning algorithms to predict the apparent permeability of the Caco-2 cell in natural products from Peru. A dataset of 1817 compounds, including experimental log Papp values and molecular descriptors, was utilized. Six QSPR models were constructed: a multiple linear regression (MLR) model, a partial least squares regression (PLS) model, a support vector machine regression (SVM) model, a random forest (RF) model, a gradient boosting machine (GBM) model, and an SVM-RF-GBM model. An evaluation of the testing set revealed that the MLR and PLS models exhibited an RMSE = 0.47 and R2 = 0.63. In contrast, the SVM, RF, and GBM models showcased an RMSE = 0.39-0.40 and R2 = 0.73-0.74. Notably, the SVM-RF-GBM model demonstrated superior performance, with an RMSE = 0.38 and R2 = 0.76. The model predicted log Papp values for 502 natural products falling within the applicability domain, with 68.9% (n = 346) showing high permeability, suggesting the potential for intestinal absorption. Additionally, we categorized the natural products into six metabolic pathways and assessed their drug-likeness. Our results provide insights into the potential intestinal absorption of natural products in Peru, thus facilitating drug development and pharmaceutical discovery efforts.

Characterization, hazard identification, and risk assessment of volatile organic compounds in Poly(butylene adipate-co-terephthalate)-based food contact articles

The chemical safety of poly (butylene adipate-co-terephthalate) (PBAT) based food contact articles (FCAs) has aroused increasing toxicological concerns in recent years, but the chemical characterization and associated risk assessment still remain inadequate as it fails to elucidate the distribution pattern and discern the potential genotoxic and carcinogenic hazards of the identified substances. Herein, the volatile organic compounds (VOCs) in 50 batches of PBAT-based FCAs of representative categories and 10 batches of PLA and PBAT pellets were characterized, by which 237 VOCs of 10 chemical categories were identified and exhibited characteristic distribution patterns in the chemical spaces derived from their molecular descriptors. Chemical hazards associated with the identified VOCs were discerned by a hazard-driven classification scheme integrating hazard-related knowledge from multiple publicly available sources, and 34 VOCs were found to bear genotoxic or carcinogenic hazards and to feature higher average molecular weight than the other VOCs. Finally, the Risk and hazard quotient (HQ) calculated as the metrics of risk suggested that all identified VOCs posed acceptable risks (Risk<10–4 or HQ < 1), whereas oxolane, butyrolactone, N,N-dimethylacetamide, 2-butoxyethanol, benzyl alcohol, and 1,2,3-trichloropropane posed non-negligible (Risk>10–6) genotoxic or carcinogenic risk and thus should be of prioritized concern to promote the chemical safety of PBAT-based FCAs.

Predicting ADMET Properties from Molecule SMILE: A Bottom-Up Approach Using Attention-Based Graph Neural Networks.

Understanding the pharmacokinetics, safety and efficacy of candidate drugs is crucial for their success. One key aspect is the characterization of absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, which require early assessment in the drug discovery and development process. This study aims to present an innovative approach for predicting ADMET properties using attention-based graph neural networks (GNNs). The model utilizes a graph-based representation of molecules directly derived from Simplified Molecular Input Line Entry System (SMILE) notation. Information is processed sequentially, from substructures to the whole molecule, employing a bottom-up approach. The developed GNN is tested and compared with existing approaches using six benchmark datasets and by encompassing regression (lipophilicity and aqueous solubility) and classification (CYP2C9, CYP2C19, CYP2D6 and CYP3A4 inhibition) tasks. Results show the effectiveness of our model, which bypasses the computationally expensive retrieval and selection of molecular descriptors. This approach provides a valuable tool for high-throughput screening, facilitating early assessment of ADMET properties and enhancing the likelihood of drug success in the development pipeline.