Molecular Descriptors Research Articles

Machine learning assisted prediction of absorption maxima in cyclohexene: A comparison using molecular descriptors and fingerprints

Light absorption plays important role in different photovoltaics devices. Easy and fast prediction of absorption properties is essential for fast screening of efficient materials. In our pursuit of identifying the optimal model for predicting absorption maxima, we systematically evaluated over 40 machine learning models, employing both molecular descriptors and fingerprints as input features. Notably, models trained on molecular descriptors demonstrated superior predictive capabilities as compared to those relying on molecular fingerprints. This not only showcased the efficacy of molecular descriptors but also highlighted the potential of these models as rapid and efficient alternatives to the density functional theory (DFT) based approaches. The use of machine learning models based on molecular descriptors introduced a level of simplicity and speed in predictions, surpassing the computational demands associated with traditional DFT-based methods. Our introduced framework that is based on machine learning, offers a valuable tool for the easy and fast prediction of properties.

Assessing solubility of meloxicam in age-specific gastric and intestinal media relevant to adults and pediatric populations: implications for optimizing dosing in patients for postoperative pain

BackgroundOral dose formulations must be soluble in gastrointestinal fluids for systemic absorption. The solubility of meloxicam was determined in 16 different age-specific simulated gastric and intestinal media that mirrored the microenvironments in pediatrics and adults.MethodsThe solubility of meloxicam in the 16 different age-specific simulated gastric and intestinal biorelevant media was assessed using the standard US pharmacopeial method. The molecular descriptors of meloxicam were used to assess its intestinal permeability.ResultsMeloxicam exhibited low solubility in the age-specific simulated gastric media for fasted and fed states and in pediatrics and adults. Similarly, meloxicam exhibited low solubility in the age-specific simulated media that mirrored neonates fed cow milk-based formula. On the other hand, meloxicam exhibited high solubility in the rest of the age-specific pediatric and adult intestinal media that simulated the fasted and fed states. The pediatric-to-adult solubility ratios were outside the 80–125% range in 7 (58.3%) and was borderline in 1 (8.3%) out of the 12 calculated ratios. These findings indicated that the solubility of meloxicam showed clinically significant differences in 8 (66.7%) of the compared media.ConclusionMeloxicam exhibited low solubility in the age-specific simulated gastric media and high solubility in the simulated intestinal media for adults and pediatrics. Moreover, the pediatric-to-adult solubility ratios may have clinically significant implications. These differences can be translated into a higher likelihood of failing to demonstrate bioequivalence of different formulations containing meloxicam and variabilities in the performance of these formulations.

Quantitative structure-property relationship for gas-chromatographic retention indices of volatile organic compounds in colored-quinoa seeds

This study outlines the development of a predictive in silico model for the retention index (RI) of 61 volatile organic compounds (VOCs) present in diverse colored-quinoa seeds. The analysis was performed using gas chromatography-ion mobility spectrometry (GC-IMS) with the FS-SE-54-CB-1 capillary column. The chemical information in the database was carefully curated, and the molecular geometries of the VOCs were properly optimized to calculate diverse molecular descriptors. The database was then divided into a training set (48 compounds) and a test set (13 molecules). For model development, unsupervised and supervised chemometric approaches were integrated to obtain a four-descriptor quantitative structure-property relationship (QSPR). The negligible differences in the coefficient of determination and residual standard deviation for the training set (R2 = 0.957, s = 34.16) and test set (R2 = 0.954, s = 35.46) indicate a stable and predictive model. The QSPR model was also subjected to cross-validation using diverse strategies, along with an applicability domain assessment. A thorough explanation of the mechanistic effects of these descriptors in predicting the RI is provided, ensuring the model accomplished all the criteria defined by the Organization for Economic Co-operation and Development. This model is simple and can assist chemists working on the analysis of VOCs in quinoa seeds using GC-IMS and to gain a deeper understanding of the molecular mechanisms involved in the retention index phenomenon. This QSPR model was also applied as a tool to predict the retention index of an external set of 89 new VOCs identified in quinoa seeds through GC-IMS analysis.

Interpretable and Physicochemical-Intuitive Deep Learning Approach for the Design of Thermal Resistance of Energetic Compounds.

Thermal resistance of energetic materials is critical due to its impact on safety and sustainability. However, developing predictive models remains challenging because of data scarcity and limited insights into quantitative structure-property relationships. In this work, a deep learning framework, named EM-thermo, was proposed to address these challenges. A data set comprising 5029 CHNO compounds, including 976 energetic compounds, was constructed to facilitate this study. EM-thermo employs molecular graphs and direct message-passing neural networks to capture structural features and predict thermal resistance. Using transfer learning, the model achieves an accuracy of approximately 97% for predicting the thermal-resistance property (decomposition temperatures above 573.15 K) in energetic compounds. The involvement of molecular descriptors improved model prediction. These findings suggest that EM-thermo is effective for correlating thermal resistance from the atom and covalent bond level, offering a promising tool for advancing molecular design and discovery in the field of energetic compounds.

Synthesis, modeling, bioactivity, docking, binding energy analysis of new effective piperazine derivatives as effective drug candidates for treatment of Parkinson’s disease

In this study, density functional theory (DFT) calculations were initially conducted to explore the molecular structure, the highest occupied molecular orbital (HOMO), the lowest unoccupied molecular orbital (LUMO), LUMO-HOMO energies, as well as the excited states of new piperazine derivatives using the CAM-B3LYP/Def2TZV level of theory. Detailed quantum molecular descriptors for the compounds, including ionization potential (IP), electron affinities (EA), hardness (η), softness (S), electronegativity (μ), electrophilic index (ω), electron-donating power (ω−), electron accepting power (ω+), and energy gap (Eg), were calculated. The chemical properties and bioactivity of these compounds, such as molecular weight, lipophilicity, number of hydrogen bond (HB) donors and acceptors, molar refractivity, polarity, solubility, flexibility and saturation, were also investigated. Subsequently, a molecular docking study was performed to identify effective compounds that could act as strong inhibitors against Parkinson’s Disease. The Michaelis-Menten constant (Km) was determined using the CAM-B3LYP/Def2TZV level of theory. The binding energy between the protein with ID:5CGJ and the organic compounds demonstrated excellent binding affinity. Therefore, the structures: 4-(4-methylpiperazin-1-yl)-4-oxo-N-(4-sulfamoylphenyl)butanamide (21), N-(4-((4-methylpiperazin-1-yl)methyl)phenyl)-4-(morpholinomethyl) benzamide (22), cаn bе usеd for potential application аgаinst Parkinson’s Disease.

Compatibility analysis of waste polymer recycling in asphalt binder using molecular descriptor and graph neural network

Recycling of waste polymers for valuable use is important for circular economy and environmental sustainability. This study introduces a novel approach to evaluating the compatibility between waste polymers and asphalt binders using advanced molecular representation models. The solubility parameters of waste polymers were predicted using traditional machine learning (ML) models and geometry-enhanced graph neural network (GeoGNN), respectively. The compatibility index was then calculated based on the absolute difference between the solubility parameters of polymers and asphalt. Results indicate that GeoGNN outperforms traditional ML and other GNN models due to its superior ability to capture complex spatial structures. The study also identifies key molecular descriptors that significantly influence solubility parameters of waste polymers. Given the variability in asphalt binder composition, the most compatible waste polymers differ across binders, making the data-driven approach especially valuable. The GeoGNN model greatly enhances the ability to assess compatibility in the polymer-asphalt system. This complements experimental techniques by integrating geometric information to analyze molecular features uncovering the structure-property relationship of material.

Quantum Chemical (QC) Calculations and Linear Solvation Energy Relationships (LSER): Hydrogen-Bonding Calculations with New QC-LSER Molecular Descriptors

Quantum Chemical (QC) Calculations and Linear Solvation Energy Relationships (LSER): Hydrogen-Bonding Calculations with New QC-LSER Molecular Descriptors

Understanding the adsorption performance of hetero-nanocages (C12-B6N6, C12-Al6N6, and B6N6-Al6N6) towards hydroxyurea anticancer drug: a comprehensive study using DFT.

Cancer is a paramount health challenge to global health, which forms tumors that can invade nearby tissues and spread to neighboring cells. Recently, nanotechnology has been used to control the growth of cancer, in which anticancer drugs are delivered to cancerous cells via nanoparticles without damaging healthy tissues. In this study, DFT investigations were carried out to examine the adsorption behavior of C24, B12N12, and Al12N12 nanocages as well as their heterostructures C12-B6N6, C12-Al6N6, and B6N6-Al6N6 towards the hydroxyurea (HU) anticancer drug. In this regard, adsorption energy, interaction distance between the drug and nanocages, charge transfer, energy gap, dipole moment, quantum molecular descriptors, work function, and COSMO surface analysis were analyzed to understand their adsorption performance. Findings demonstrate that the adsorption energies of two hetero-nanocages on their hexagonal (SH) and tetragonal (ST) sites are favorable for the drug delivery process. The computed adsorption energy of B6N6-Al6N6 of the ST/AlN site is 183.59 kJ mol-1, which is higher than that of the C12-Al6N6 nanocage, including minimum adsorption distances. Negative adsorption energy with low adsorption distances implies an attractive interaction between the drug and nanocages. During the interaction, a significant amount of charge is transferred between the drug and nanocages. Furthermore, for both complexes, larger dipole moments were observed in water media compared to gas media. From DOS spectra, prominent peaks were found in the Fermi level after adsorption of HU on the nanocages, implying the reduction of the energy gap. Noticeable overlaps between the PDOS spectra of the nanocages and HU's close contact atom demonstrate the formation of chemical bonds between two specific atoms. Therefore, it can be concluded that among the nanocages, C12-Al6N6 and B6N6-Al6N6 may be suitable carriers for HU drug.

Exploring the Antibacterial and Antiparasitic Activity of Phenylaminonaphthoquinones-Green Synthesis, Biological Evaluation and Computational Study.

Organic compounds with antibacterial and antiparasitic properties are gaining significance for biomedical applications. This study focuses on the solvent-free synthesis (green synthesis) of 1,4-naphthoquinone or 2,3-dichloro-1,4-naphthoquinone with different phenylamines using silica gel as an acid solid support. The study also includes in silico PASS predictions and the discovery of antibacterial and antiparasitic properties of phenylaminonaphthoquinone derivatives 1-12, which can be further applied in drug discovery and development. These activities were discussed in terms of molecular descriptors such as hydrophobicity, molar refractivity, and half-wave potentials. The in vitro antimicrobial potential of the synthesized compounds 1-12 was evaluated against a panel of six bacterial strains (three Gram-positive: Staphylococcus aureus, Proteus mirabilis, and Enterococcus faecalis; and three Gram-negative bacteria: Escherichia coli, Salmonella typhimurium, and Klebsiella pneumoniae). Six compounds (1, 3, 5, 7, 10, and 11) showed better activity toward S. aureus with MIC values between 3.2 and 5.7 μg/mL compared to cefazolin (MIC = 4.2 μg/mL) and cefotaxime (MIC = 8.9 μg/mL), two cephalosporin antibiotics. Regarding in vitro antiplasmodial activity, compounds 1 and 3 were the most active against the Plasmodium falciparum strain 3D7 (chloroquine-sensitive), displaying IC50 values of 0.16 and 0.0049 μg/mL, respectively, compared to chloroquine (0.33 μg/mL). In strain FCR-3 (chloroquine-resistant), most of the compounds showed good activity, with compounds 3 (0.12 μg/mL) and 11 (0.55 μg/mL) being particularly noteworthy. Additionally, docking studies were used to better rationalize the action and prediction of the binding modes of these compounds. Finally, absorption, distribution, metabolism, excretion, and toxicity (ADMET) predictions were performed.

Using artificial neural networks to elucidate retention interactions on stationary phases with amine moieties dedicated to supercritical fluid chromatography

The retention mechanisms of small molecules on a wide range of stationary phases using different mobile phase compositions are one of the most discussed topics in current state-of-the-art supercritical fluid chromatography (SFC). Although various theories have been published on retention behavior, prediction, and stability over time, an in-depth insight into the mechanisms is still lacking. Our study focused on the description of the retention mechanisms on columns with amine moieties, i.e., 2-ethylpyridine, 2-picolylamine, 1-aminoanthracene, and diethylamine, dedicated to SFC, using three different compositions of organic modifier, i.e., methanol, methanol + 10 mmol/L NH3, and methanol + 2 % H2O, in CO2. An extensive set of analytes characterized by more than 200 molecular descriptors was used. The importance of the effect of each molecular descriptor on the retention behavior was determined by artificial neural networks. Thus, the descriptors increasing and/or decreasing retention on individual columns using different mobile phase compositions were defined. The data collected over one year of column use were statistically evaluated to describe the retention behavior of small molecules on the tested stationary phases using three organic modifiers and to describe the changes in retention behavior over time.As the development of the SFC method can be laborious and time-consuming, this publication offers a detailed description of the interactions taking place between the analyte and the SFC stationary phase. Understanding these fundamental processes will enable faster development of SFC methods using quality-by-design principles.

Machine learning analysis of a large set of homopolymers to predict glass transition temperatures

Glass transition temperature of polymers, Tg, is an important thermophysical property, which sometimes can be difficult to measure experimentally. In this regard, data-driven machine learning approaches are important alternatives to assess Tg values, in a high-throughput way. In this study, a large dataset of more than 900 polymers with reported glass transition temperature (Tg) was assembled from various public sources in order to develop a predictive model depicting the structure-property relationships. The collected dataset was curated, explored via cluster analysis, and then split into training and test sets for validation purposes and then polymer structures characterized by molecular descriptors. To find the models, several machine learning techniques, including multiple linear regression (MLR), k-nearest neighbor (k-NN), support vector machine (SVM), random forest (RF), gaussian processes for regression (GPR), and multi-layer perceptron (MLP) were explored. As result, a model with the subset of 15 descriptors accurately predicting the glass transition temperatures was developed. The electronic effect indices were determined to be important properties that positively contribute to the Tg values. The SVM-based model showed the best performance with determination coefficients (R2) of 0.813 and 0.770, for training and test sets, respectively. Also, the SVM model showed the lowest estimation error, RMSE = 0.062. In addition, the developed structure-property model was implemented as a web app to be used as an online computational tool to design and evaluate new homopolymers with desired glass transition profiles.

Understanding the mechanism of microplastic-associated antibiotic resistance genes in aquatic ecosystems: Insights from metagenomic analyses and machine learning

The pervasive presence of microplastics (MPs) in aquatic systems facilitates the transmission of antibiotic resistance genes (ARGs), thereby posing risks to ecosystems and human well-being. However, owing to variations in environmental backgrounds and the limited scope of research subjects, studies on ARGs in MPs lack unified conclusions, particularly regarding whether different types of MPs selectively promote ARG enrichment. Analysing large-scale datasets can better encompass broad spatiotemporal scales and diverse samples, facilitating a more extensive exploration of the complex ecological relationships between MPs and ARGs. The present study integrated existing metagenomic datasets to conduct a comprehensive risk assessment and comparative analysis of resistance groups across various MPs. In addition, we endeavoured to elucidate potential associations between ARGs and bacterial taxa, as well as MP structural features, using machine learning (ML) methods. The findings of our research highlight the pivotal role of MP type in shaping plastispheres, accounting for 9.56 % of the biotic variation (Adonis index) and explaining 18.59 % of the ARG variance. Compared to conventional MPs, biodegradable MPs, such as polyhydroxyalkanoates (PHA) and polylactic acid (PLA), exhibit lower species uniformity and diversity but pose a higher risk of ARG occurrence. These ML approaches effectively forecasted ARG abundance by using the bacterial taxa and molecular structure descriptors (MDs) of MPs (average R2tra = 0.882, R2test = 0.759). Feature analysis showed that MDs associated with lipophilicity, solubility, toxicity, and surface potential significantly influenced the relative abundance of ARGs in the plastispheres. The interpretable multiple linear regression (MLR) model, particularly notable, elucidated a linear relationship between bacterial genera and ARGs, offering promise for identifying potential ARG hosts. This study offers novel insights into ARG dynamics and ecological risks within aquatic plastispheres, highlighting the importance of comprehensive MP monitoring initiatives.

Machine Learning-Based Toxicological Modeling for Screening Environmental Obesogens.

The emerging presence of environmental obesogens, chemicals that disrupt energy balance and contribute to adipogenesis and obesity, has become a major public health challenge. Molecular initiating events (MIEs) describe biological outcomes resulting from chemical interactions with biomolecules. Machine learning models based on MIEs can predict complex toxic end points due to chemical exposure and improve the interpretability of models. In this study, a system was constructed that integrated six MIEs associated with adipogenesis and obesity. This system showed high accuracy in external validation, with an area under the receiver operating characteristic curve of 0.78. Molecular hydrophobicity (SlogP_VSA) and direct electrostatic interactions (PEOE_VSA) were identified as the two most critical molecular descriptors representing the obesogenic potential of chemicals. This system was further used to predict the obesogenic effects of chemicals on the candidate list of substances of very high concern (SVHCs). Results from 3T3-L1 adipogenesis assays verified that the system correctly predicted obesogenic or nonobesogenic effects of 10 of the 12 SVHCs tested, and identified four novel potential obesogens, including 2-benzotriazol-2-yl-4,6-ditert-butylphenol (UV-320), 4-(1,1,5-trimethylhexyl)phenol (p262-NP), 2-[4-(1,1,3,3-tetramethylbutyl)phenoxy]ethanol (OP1EO) and endosulfan. These validation data suggest that the screening system has good performance in adipogenic prediction.

Transfer learning for molecular property predictions from small datasets

Machine learning has emerged as a new tool in chemistry to bypass expensive experiments or quantum-chemical calculations, for example, in high-throughput screening applications. However, many machine learning studies rely on small datasets, making it difficult to efficiently implement powerful deep learning architectures such as message passing neural networks. In this study, we benchmark common machine learning models for the prediction of molecular properties on two small datasets, for which the best results are obtained with the message passing neural network PaiNN as well as SOAP molecular descriptors concatenated to a set of simple molecular descriptors tailored to gradient boosting with regression trees. To further improve the predictive capabilities of PaiNN, we present a transfer learning strategy that uses large datasets to pre-train the respective models and allows us to obtain more accurate models after fine-tuning on the original datasets. The pre-training labels are obtained from computationally cheap ab initio or semi-empirical models, and both datasets are normalized to mean zero and standard deviation one to align the labels’ distributions. This study covers two small chemistry datasets, the Harvard Organic Photovoltaics dataset (HOPV, HOMO–LUMO-gaps), for which excellent results are obtained, and the FreeSolv dataset (solvation energies), where this method is less successful, probably due to a complex underlying learning task and the dissimilar methods used to obtain pre-training and fine-tuning labels. Finally, we find that for the HOPV dataset, the final training results do not improve monotonically with the size of the pre-training dataset, but pre-training with fewer data points can lead to more biased pre-trained models and higher accuracy after fine-tuning.

Random Forest‐Based Prediction of Enantioselectivity in Thiol Addition to Imines Catalyzed by Chiral Phosphoric Acids

AbstractThis paper, for the first time, reports calculating molecular descripts from the combinations of catalysts and substrates, for enantioselectivity predictions in chiral phosphoric acid–catalyzed thiol addition to N‐acylimines. Ten molecular descriptors, together with random forest (RF) algorithm, were used to establish the quantitative structure–selectivity relationship (QSSR) model (RF Model I), yielding correlation coefficient of R=0.982, root mean square (rms) error of 0.604 kJ/mol (for 800 enantioselectivity ΔΔG from 32 catalysts in the training set), R=0.922, rms=1.075 kJ/mol (for 275 enantioselectivity ΔΔG from 11 new catalysts in the test set). The RF Model I was proved to be more accurate in enantioselectivity predictions, compared with the QSSR models reported in the literature and with the RF Model II and RF Model III in this work. Moreover, the RF Model I was based on simple Dragon descriptors and capable of extrapolative prediction for new catalysts. Therefore, it is reasonable deriving molecular descriptors from the combinations of catalysts and substrates (thiols and imines). This work provides a new method for the predictions of enantioselectivity ΔΔG.

Molecular descriptors of symmetrically configured carbon nanocones via quotient graph technique

Carbon nanocones, characterized by planar networks primarily consisting of hexagonal carbon faces and a few pentagons in the core, hold significant promise in various fields such as electronics, medicine, and material research due to their exceptional properties. This study exclusively investigates the topological descriptors of symmetrically configured carbon nanocones, specifically nanocones with two and three pentagonal core respectively. By modeling these structures as chemical graphs and employing molecular descriptors, we aim to provide a quantitative analysis of the structures without laboratory experimentation. The quotient graph approach is employed to determine the distance based descriptors, namely wiener, szeged and Padmakar-Ivan indices for these symmetrically configured carbon nanocones. The analytically closed formulas determined could be useful in quantitative structure-property-activity relationship studies, which helps in predicting the physico-chemical properties of the underlying structures. Thus the findings in this work not only deepen our understanding of symmetric carbon nanocones but also pave the way for significant advancements in nanotechnology and materials science.

Using the Super-learner to Predict the Chemical Acute Toxicity on Rats

With the rapid increase in the number of commercial chemicals, testing methods regarding on median lethal dose (LD50) relying animal experiments face challenges such as high costs and ethical concerns. Classical quantitative structure-activity relationship models relying on single algorithm always lack interpretability and precision, given the complexity of the mechanisms underlying acute toxicity. To address these issues, this study has developed a predictive framework using an ensemble learning model based on Super-learner. Particularly, we first obtained LD50 data for 9,843 compounds and constructed 16 meta models using 4 molecular descriptors and machine learning algorithms. The Super-learner model performed well, achieving R² values of 0.61 and 0.64 in five-fold cross-validation and test sets, respectively, with corresponding root mean square errors of 0.55 and 0.64, significantly outperforming the results of individual model. Additionally, we incorporated data filtering and applicability domain methods, which demonstrated that the Super-learner can mitigate the impact of dataset noise to some extent. The model achieved an R² of 0.76 within an applicability domain, ensuring prediction accuracy within the chemical space. Compared to previous studies, the model developed here using Super-learner generally achieved better performance across a larger applicability domain. Finally, we has launched an online tool (http://sltox.hhra.net), allowing users to quickly predict LD50 of compounds, greatly simplifying the chemical safety assessment process. This study not only provides an effective and cost-efficient method for predicting chemical toxicity but also offers technical support and data for risk assessments of chemicals.

A phase separation prediction model for CO2 capture by biphasic amine absorbents

The biphasic amine absorbent upon absorbing CO2, separates into CO2 poor and CO2 rich phases, making it a third-generation chemical absorbent with energy-saving potential. However, the current method of constructing biphasic amine absorbents involves screening phase separation agents through a large number of experiments, which involves significant amount of trial and error. In this study, a phase separation prediction model was built for CO2 capture by biphasic amine absorbents for quick screening of phase separation agents. The phase separation of CO2 absorption by organic amine 2-amino-2-methyl-1-propanol (AMP) using 34 phase separation agents was statistically analyzed based on 14 molecular descriptors of the solvents. The factors influencing the phase separation were thoroughly explored. It is proposed that the relative dielectric constant or molecular polarity index can replace the traditional polarity evaluation index, dipole moment, to better describe the phase separation situation. Additionally, it is found that the ability of phase separation agents to form hydrogen bonds, particularly as a hydrogen bond donor is the determining factor for phase separation to occur. Ultimately, the preferred molecular descriptors are used as predictors, leading to the development of a prediction model for phase separation that achieves high-accuracy predictions solely through theoretical calculation parameters. The model considerably reducing the construction costs of biphasic amine absorbents.

1,5-Disubstituted-1,2,3-Triazoles as GABA analogues: Synthesis, QSAR and biological evaluation as Pseudomonas fluorescens GABA-AT inhibitors

We report the synthesis of a new series of γ-aminobutyric acid (GABA) analogues as possible GABA-AT inhibitors, where the nitrogen at the Ƴ-position is contained in a 1,5-disubstituted-1,2,3-triazole ring system. The triazole ring system was assembled by the Ru(I)-catalyzed alkyne-azide 1,3-dipolar cycloaddition (RuACC) protocol. GABA-AT inhibition activity was predicted by means of a QSAR model that was constructed in this work. Different molecular descriptors families were employed for its construction, like steric type (Wap and PW5), solubility type (ΔPSA), and electronic type (ΔTriazole3). Additionally, a similarity analysis based on the electronic profile of 1,5-triazole and 1,4-triazole GABA analogues was done to support the QSAR predictions. Finally, the three best candidates were biologically tested as Pseudomonas fluorescens GABA-AT inhibitors showing an inhibition percent compared to FDA approved inhibitors.

In silico models for the screening of human transthyretin disruptors

The use of New Approach Methodologies (NAMs), such as Quantitative Structure-Activity Relationship (QSAR) models, is highly recommended by international regulations to speed up hazard and risk assessment of Endocrine Disruptors, which are known to be linked to a wide spectrum of severe diseases on humans and wildlife. A very sensitive target for these chemicals is the thyroid hormone system, which plays a key role in regulating metabolic and cognitive functions. Several chemicals have been demonstrated to compete with the thyroid hormone thyroxine (T4) for binding to human thyroid hormone distributor protein transthyretin (hTTR). In this work, we generated three new datasets composed by T4-hTTR competing potencies of more than 200 heterogeneous chemicals measured by three different in vitro assays. These datasets were used for the development of new regression QSAR models. The best models were thoroughly validated by internal and external validation procedures. The mechanistic interpretation of the selected molecular descriptors provided information on structural features which are relevant to characterise hTTR binders, such as the presence of hydroxylated and halogenated aromatic rings. PCA analysis was used to rank the studied chemicals according to their increasing T4-hTTR competing potency. Hydroxylated and halogenated bicyclic aromatic compounds are ranked as the strongest hTTR binders. The new QSARs are useful to screen potential Thyroid Hormone System-Disrupting Chemicals (THSDCs), and to support the identification of sustainable alternatives to hazardous chemicals.