Molecular Fingerprints Research Articles

Innovative Mamba and graph transformer framework for superior protein-ligand affinity prediction

Accurate measurement of the affinity between drug targets is of great importance in the field of drug discovery, as it provides significant information about drug action. However, the diverse representations of compounds and proteins increase the difficulty of this task. Recently, numerous studies have explored the application of various deep learning methods and architectures in Drug-Target Affinity (DTA) prediction tasks, continuously improving performance. In this work, we present a multimodal late-stage fusion prediction model called MGDTA that combines the pioneering Mamba architecture with a hybrid model architecture based on graph transformers and transformers (GTT). For molecular compounds, structural characterisation is extracted using GTT layers. Additionally, a feature integration module is utilized to obtain molecular features from derived molecular fingerprints, culminating in a unified representation of drug molecules through a feature integration module. For proteins, we adopt the state-space model (SSM) module Mamba to learn feature representations from target protein sequences. This represents a preliminary exploration of the Mamba module in protein sequence feature extraction in DTA tasks. Finally, we predict the results by integrating the feature representations of both molecules and proteins. Our results indicate that the proposed model achieves superior performance on benchmark datasets, validating the feasibility of using the Mamba architecture for DTA tasks. Furthermore, sequence feature visualization demonstrates the capability of the Mamba block to focus on binding site residues within drug targets.

Standardized Electric-Field-Resolved Molecular Fingerprinting.

Field-resolved infrared spectroscopy (FRS) of impulsively excited molecular vibrations can surpass the sensitivity of conventional time-integrating spectroscopies, owing to a temporal separation of the molecular signal from the noisy excitation. However, the resonant response carrying the molecular signal of interest depends on both the amplitude and phase of the excitation, which can vary over time and across different instruments. To date, this has compromised the accuracy with which FRS measurements could be compared, which is a crucial factor for practical applications. Here, we utilize a data processing procedure that overcomes this shortcoming while preserving the sensitivity of FRS. We validate the approach for aqueous solutions of molecules. The employed approach is compatible with established processing and evaluation methods for the analysis of infrared spectra and can be applied to existing spectra from databases, facilitating the spread of FRS to new molecular analytical applications.

Long-wavelength infrared upconversion time-stretch spectroscopy

High-speed spectroscopy in the molecular fingerprint spectral region (≈6–12 μm) is essential for the detection of ultrafast molecular dynamic processes, rapid combustion analysis, and biological diagnostics. However, ultrafast spectroscopy in the long-wavelength infrared (LWIR) region remains a challenge due to the limitations of laser sources and the lack of ultrafast and sensitive detectors in this wavelength region. Here, we demonstrate broadband LWIR time-stretch spectroscopy, which can realize a single-shot high-speed spectral measurement in a 8–10 μm region, by combining the LWIR femtosecond (fs) light generation and upconversion time-stretching detection with specific dispersive fiber. Broadband tunable fs light generated in the 8–10 μm region is upconverted to the 1.1–1.2 μm near-infrared wavelength via difference-frequency generation with the 1 μm chirped pump pulse. Time-stretch detection of the upconverted light can then be realized by adopting dispersion shifted fiber, which has a superior dispersion-to-loss ratio in the 1.1–1.2 μm wavelength region, as the dispersive medium. As a result, we experimentally demonstrate LWIR time-stretch spectroscopy in the 8–10 μm region with a spectral resolution of 1.07 cm−1, at a rate of 200 kSpectra s−1, which is only limited by the repetition rate of the 1 μm pump source. The demonstration of high-speed time-stretch spectroscopy in the LWIR region would open the possibility in exploring the transient dynamics of molecular fingerprint spectroscopy.

Development of a Plasma-Based Nanosensor Platform for Triage of Neurological Tumors

Neurological tumors have diverse etiologies and treatments. Presently, diagnostic technologies such as MRI or CT scans are limited by poor resolution, and thus imprecise identification of tumor classes. As a result, diagnostic methods rely on highly invasive intracranial biopsies for conclusive tumor identification. Development of non-invasive methods to accurately identify tumor class may lead to earlier and more accurate diagnoses, enabling prompt and targeted treatment strategies. To address this unmet clinical need, we report the development a molecular-recognition-based nanosensor array that can accurately identify a molecular fingerprint of glioma and meningioma from plasma. We developed this capability using sensor arrays of chemically modified carbon nanotubes that can transduce subtle differences in the physicochemical properties of molecules in biofluids via sensitive molecular binding interactions. Using high-throughput near-infrared fluorescence spectroscopy, we screened 23 chemically modified (6,5) carbon nanotubes sensors in 324 patient plasma samples. We trained machine learning models to classify sensor responses as coming from either glioma or meningioma tumors. The best performing models were able to achieve a positive predictive value of 0.75. Sensor feature analysis revealed a set of spectral features positively associated with class differentiation. This work has the potential to aid clinical care by facilitating the early triage of different types of tumors to provide better treatment options for patients.

In situ observation of chemistry in Rydberg molecules within a Bose-Einstein-condensate

We often infer the state of systems in nature indirectly, for example, in high-energy physics by the interaction of particles with an ambient medium. We adapt this principle to energies 9 orders of magnitude smaller, to classify the final state of exotic molecules after internal conversion of their electronic state, through their interaction with an ambient quantum fluid, a Bose-Einstein condensate (BEC). The BEC is the ground state of a million bosonic atoms near zero temperature, and a single embedded ultra-long-range Rydberg molecule can coherently excite waves in this fluid, which carry telltale signatures of its dynamics. Bond lengths exceeding a micrometer allow us to observe the molecular fingerprint on the BEC , via optical microscopy. Interpreting images in comparison with simulations strongly suggests that the molecular electronic state rapidly converts from the initially excited S and D orbitals to a much more complex molecular state (called “trilobite”), marked by a maximally localized electron. This internal conversion liberates energy, such that one expects final-state particles to move rapidly through the medium, which is however ruled out by comparing experiment and simulations. The molecule thus must strongly decelerate in the medium, for which we propose a plausible mechanism. Our experiment demonstrates a medium that facilitates and records an electronic state change of embedded exotic molecules in ultracold chemistry, with sufficient sensitivity to constrain velocities of final-state particles. Published by the American Physical Society 2024

Broadband terahertz absorption and Q-switching behavior of 5-chloro-2-nitroaniline (5C2NA) crystals

Abstract Growth of transparent 5-chloro-2-nitroaniline (5C2NA) crystals was achieved through the slow evaporation technique. Analysis via X-ray diffraction (XRD) and Fourier Transform Infrared (FTIR) spectroscopy confirmed the crystalline nature of the organic crystals and revealing distinct molecular fingerprints of 5C2NA respectively. UV-VIS and Photoluminescence (PL) spectroscopy were employed to study the material's band gap and ground state absorption respectively. Thermogravimetric analysis (TGA) analysis indicated the stability of the grown crystals up to 211ºC. Dielectric measurements and Urbach plot suggested the presence of fewer defects in 5C2NA crystals. Time domain terahertz spectroscopy was utilized to observe the variation of absorption coefficient and refractive index in the terahertz frequency regime. Nonlinear optical effects such as saturable absorption (SA) and reverse saturable absorption (RSA), are pivotal in the development of all-optical logic gates. The transition between SA and RSA possesses a significant importance in optoelectronic applications. In this study, we investigate the 5C2NA crystal, revealing its ability to exhibit both SA and RSA under Z scan technique with varying pump intensities. The switching properties observed in 5C2NA can be harnessed for applications such as all-optical logic gates, rapid optical switching, optical limiting, mode storage, and more.

Inhalation Toxicity Screening of Consumer Products Chemicals using OECD Test Guideline Data-based Machine Learning Models

This study aimed to screen the inhalation toxicity of chemicals found in consumer products such as air fresheners, fragrances, and anti-fogging agents submitted to K-REACH using machine learning models. We manually curated inhalation toxicity data based on OECD test guideline 403 (Acute inhalation), 412 (Sub-acute inhalation), and 413 (Sub-chronic inhalation) for 1709 chemicals from the OECD eChemPortal database. Machine learning models were trained using ten algorithms, along with four molecular fingerprints (MACCS, Morgan, Topo, RDKit) and molecular descriptors, achieving F1 scores ranging from 51 % to 91 % in test dataset. Leveraging the high-performing models, we conducted a virtual screening of chemicals, initially applying them to data-rich chemicals generally used in occupational settings to determine the prediction uncertainty. Results showed high sensitivity (75 %) but low specificity (23 %), suggesting that our models can contribute to conservative screening of chemicals. Subsequently, we applied the models to consumer product chemicals, identifying 79 as of high concern. Most of the prioritized chemicals lacked GHS classifications related to inhalation toxicity, even though they were predicted to be used in many consumer products. This study highlights a potential regulatory blind spot concerning the inhalation risk of consumer product chemicals while also indicating the potential of artificial intelligence (AI) models to aid in prioritizing chemicals at the screening level.

Analysis of microisolated frontal cortex excitatory layer III and V pyramidal neurons reveals a neurodegenerative phenotype in individuals with Down syndrome.

We elucidated the molecular fingerprint of vulnerable excitatory neurons within select cortical lamina of individuals with Down syndrome (DS) for mechanistic understanding and therapeutic potential that also informs Alzheimer's disease (AD) pathophysiology. Frontal cortex (BA9) layer III (L3) and layer V (L5) pyramidal neurons were microisolatedfrom postmortem human DS and age- and sex-matched controls (CTR) to interrogate differentially expressed genes (DEGs) and key biological pathways relevant to neurodegenerative programs. We identified > 2300 DEGs exhibiting convergent dysregulation of gene expression in both L3 and L5 pyramidal neurons in individuals with DS versus CTR subjects. DEGs included over 100 triplicated human chromosome 21 genes in L3 and L5 neurons, demonstrating a trisomic neuronal karyotype in both laminae. In addition, thousands of other DEGs were identified, indicating gene dysregulation is not limited to trisomic genes in the aged DS brain, which we postulate is relevant to AD pathobiology. Convergent L3 and L5 DEGs highlighted pertinent biological pathways and identified key pathway-associated targets likely underlying corticocortical neurodegeneration and related cognitive decline in individuals with DS. Select key DEGs were interrogated as potential hub genes driving dysregulation, namely the triplicated DEGs amyloid precursor protein (APP) and superoxide dismutase 1 (SOD1), along with key signaling DEGs including mitogen activated protein kinase 1 and 3 (MAPK1, MAPK3) and calcium calmodulin dependent protein kinase II alpha (CAMK2A), among others. Hub DEGs determined from multiple pathway analyses identified potential therapeutic candidates for amelioration of cortical neuron dysfunction and cognitive decline in DS with translational relevance to AD.

Machine learning using fingerprints and dye design in the search of lower hole reorganization energy

Organic solar cells (OSCs) are gaining fame for their cost-effective solution processing. Machine learning is increasingly popular for material design in OSCs. In this study, molecular fingerprints are used to train over 40 machine learning models. The random forest regressor emerges as the most predictive one. 10k new dyes are generated. A pre-trained ML model is used to predict their reorganization energy values. Dyes are selected on the basis of reorganization energy, dyes with lower reorganization energy are retrained. The synthetic accessibility of chosen dyes is then analyzed. Chemical similarity analysis has indicated reasonable resemble among selected dyes.

Identification of Pristine and Protein Corona Coated Micro- and Nanoplastic Particles with a Colorimetric Sensor Array.

A colorimetric sensor array has been developed to differentiate various micro- and nanoplastic particles (MNPs), both pristine and those coated with a protein corona, in buffered water. This array utilizes five distinct cross-reactive chemo-responsive dyes, which exhibit changes in visible optical absorbance upon interaction with MNPs. Although no single dye responds exclusively to either pristine or protein-corona-coated MNPs, the collective shifts in color across all dyes create a unique molecular fingerprint for each type of MNP. This method demonstrates high sensitivity, capable of detecting MNPs of various sizes (50 nm, 100 nm, and 2 μm) and differentiating them from controls at concentrations as low as 10 ng/mL using standard chemometric techniques, ensuring accurate results without error. Additionally, the method can effectively distinguish between pristine and protein-corona-coated polystyrene MNPs. This colorimetric approach offers a rapid, cost-effective, and accurate method for monitoring MNP pollution and assessing their prior interactions with biological systems.

Predicting repurposed drugs targeting the NS3 protease of dengue virus using machine learning-based QSAR, molecular docking, and molecular dynamics simulations

ABSTRACT Dengue fever, prevalent in Southeast Asian countries, currently lacks effective pharmaceutical interventions for virus replication control. This study employs a strategy that combines machine learning (ML)-based quantitative-structure-activity relationship (QSAR), molecular docking, and molecular dynamics simulations to discover potential inhibitors of the NS3 protease of the dengue virus. We used nine molecular fingerprints from PaDEL to extract features from the NS3 protease dataset of dengue virus type 2 in the ChEMBL database. Feature selection was achieved through the low variance threshold, F-Score, and recursive feature elimination (RFE) methods. Our investigation employed three ML models – support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost) – for classifier development. Our SVM model, combined with SVM-RFE, had the best accuracy (0.866) and ROC_AUC (0.964) in the testing set. We identified potent inhibitors on the basis of the optimal classifier probabilities and docking binding affinities. SHAP and LIME analyses highlighted the significant molecular fingerprints (e.g. ExtFP69, ExtFP362, ExtFP576) involved in NS3 protease inhibitory activity. Molecular dynamics simulations indicated that amphotericin B exhibited the highest binding energy of −212 kJ/mol and formed a hydrogen bond with the critical residue Ser196. This approach enhances NS3 protease inhibitor identification and expedites the discovery of dengue therapeutics.

Construction of IRAK4 inhibitor activity prediction model based on machine learning.

Interleukin-1 receptor-associated kinase 4 (IRAK4) is a crucial serine/threonine protein kinase that belongs to the IRAK family and plays a pivotal role in Toll-like receptor (TLR) and Interleukin-1 receptor (IL-1R) signaling pathways. Due to IRAK4's significant role in immunity, inflammation, and malignancies, it has become an intriguing target for discovering and developing potent small-molecule inhibitors. Consequently, there is a pressing need for rapid and accurate prediction of IRAK4 inhibitor activity. Leveraging a comprehensive dataset encompassing activity data for 1628 IRAK4 inhibitors, we constructed a prediction model using the LightGBM algorithm and molecular fingerprints. This model achieved an R2 of 0.829, an MAE of 0.317, and an RMSE of 0.460 in independent testing. To further validate the model's generalization ability, we tested it on 90 IRAK4 inhibitors collected in 2023. Subsequently, we applied the model to predict the activity of 13,268 compounds with docking scores less than -9.503kcal/mol. These compounds were initially screened from a pool of 1.6 million molecules in the chemdiv database through high-throughput molecular docking. Among these, 259 compounds with predicted pIC50 values greater than or equal to 8.00 were identified. We then performed ADMET predictions on these selected compounds. Finally, through a rigorous screening process, we identified 34 compounds that adhere to the four complementary drug-likeness rules, making them promising candidates for further investigation. Additionally, molecular dynamics simulations confirmed the stable binding of the screened compounds to the IRAK4 protein. Overall, this work presents a machine learning model for accurate prediction of IRAK4 inhibitor activity and offers new insights for subsequent structure-guided design of novel IRAK4 inhibitors.

Evaluating ionic liquid toxicity with machine learning and structural similarity methods

Ionic liquids (ILs) have garnered significant interest owing to their distinct physicochemical traits. Nonetheless, their extensive application is curtailed by ecotoxicity concerns. This study aimed to develop a quantitative structure-activity relationship (QSAR) model for predicting the toxicity of ILs in biological cells. Toxicity data of ILs on leukemia rat cell line IPC-81, Escherichia coli (E. coli), and acetylcholinesterase (AChE) were collected from open-source databases, and two integrated models, random forest (RF) and gradient boosted decision tree (GBDT), were used to train the data. The molecular structures of the ILs were represented by three different methods, namely molecular descriptor (MD), molecular fingerprint (MF), and molecular identifier (MI), respectively. The Tanimoto similarity coefficients indicate that MD has a stronger ability to recognize structural similarity. Statistical metrics of model performance showed that the two models (MD-RF and MD-GBDT) with MD as an input feature performed better in the three datasets. The application of the SHapley Additive exPlanations (SHAP) method explains the importance of different features. Specifically, reducing the carbon chain length and the number of fluorine atoms in the structure of ILs can effectively reduce their toxic effects on biological cells. This study employs machine learning to grasp better how the structure of ILs relates to inhibiting biotoxicity, offering insights for crafting safer, eco-friendly IL designs.

Diagnostic application in streptozotocin-induced diabetic retinopathy rats: A study based on Raman spectroscopy and machine learning.

Vision impairment caused by diabetic retinopathy (DR) is often irreversible, making early-stage diagnosis imperative. Raman spectroscopy emerges as a powerful tool, capable of providing molecular fingerprints of tissues. This study employs RS to detect ex vivo retinal tissue from diabetic rats at various stages of the disease. Transmission electron microscopy was utilized to reveal the ultrastructural changes in retinal tissue. Following spectral preprocessing of the acquired data, the random forest and orthogonal partial least squares-discriminant analysis algorithms were employed for spectral data analysis. The entirety of Raman spectra and all annotated bands accurately and distinctly differentiate all animal groups, and can identify significant molecules from the spectral data. Bands at 524, 1335, 543, and 435 cm-1 were found to be associated with the preproliferative phase of DR. Bands at 1045 and 1335 cm-1 were found to be associated with early stages of DR.

Integrating Prior Chemical Knowledge into the Graph Transformer Network to Predict the Stability Constants of Chelating Agents and Metal Ions.

The latest advancements in nuclear medicine indicate that radioactive isotopes and associated metal chelators play crucial roles in the diagnosis and treatment of diseases. The development of metal chelators mainly relies on traditional trial-and-error methods, lacking rational guidance and design. In this study, we propose the structure-aware transformer (SAT) combined with molecular fingerprint (SATCMF), a novel graph transformer network framework that incorporates prior chemical knowledge to construct coordination edges and learns the interactions between chelating agents and metal ions. SATCMF is trained on stability data collected from metal ion-ligand complexes, leveraging the SAT network to extract structural features relevant to the binding of ligands with metal ions. It further integrates molecular fingerprint features to refine the prediction of the stability constants of the chelating agents and metal ions. The experimental results on benchmark data set demonstrate that SATCMF achieves state-of-the-art performance based on four different graph neural network architectures. Additionally, visualizing the learned molecular attention distribution provides interpretable insights from the prediction results, offering valuable guidance for the development of novel metal chelators.

Fully Flexible Molecular Alignment Enables Accurate Ligand Structure Modeling.

Accurate protein-ligand binding poses are the prerequisites of structure-based binding affinity prediction and provide the structural basis for in-depth lead optimization in small molecule drug design. However, it is challenging to provide reasonable predictions of binding poses for different molecules due to the complexity and diversity of the chemical space of small molecules. Similarity-based molecular alignment techniques can effectively narrow the search range, as structurally similar molecules are likely to have similar binding modes, with higher similarity usually correlated to higher success rates. However, molecular similarity is not consistently high because molecules often require changes to achieve specific purposes, leading to reduced alignment precision. To address this issue, we propose a new alignment method─Z-align. This method uses topological structural information as a criterion for evaluating similarity, reducing the reliance on molecular fingerprint similarity. Our method has achieved success rates significantly higher than those of other methods at moderate levels of similarity. Additionally, our approach can comprehensively and flexibly optimize bond lengths and angles of molecules, maintaining a high accuracy even when dealing with larger molecules. Consequently, our proposed solution helps in achieving more accurate binding poses in protein-ligand docking problems, facilitating the development of small molecule drugs. Z-align is freely available as a web server at https://cloud.zelixir.com/zalign/home.

Prediction of PFAS bioaccumulation in different plant tissues with machine learning models based on molecular fingerprints

Due to the wastewater irrigation or biosolid application, per- and polyfluoroalkyl substances (PFASs) have been widely detected in agriculture soil and hence crops or vegetables. Consumption of contaminated crops and vegetables is considered as an important route of human exposure to PFASs. Machine learning (ML) models have been developed to predict PFAS uptake by plants with majority focus on roots. However, ML models for predicting accumulation of PFASs in above ground edible tissues have yet to be investigated. In this study, 811 data points covering 22 PFASs represented by molecular fingerprints and 5 plant categories (namely the root class, leaf class, cereals, legumes, and fruits) were used for model development. The Extreme Gradient Boosting (XGB) model demonstrated the most favorable performance to predict the bioaccumulation factors (BAFs) in all the 4 plant tissues (namely root, leaf, stem, and fruit) achieving coefficients of determination R2 as 0.82–0.93. Feature importance analysis showed that the top influential factors for BAFs varied among different plant tissues, indicating that model developed for root concentration prediction may not be feasible for above ground parts. The XGB model's performance was further demonstrated by comparing with data from pot experiments measuring BAFs of 12 PFASs in lettuce. The correlation between predicted and measured results was favorable for BAFs in both lettuce roots and leaves with R2 values of 0.76 and 0.81. This study developed a robust approach to comprehensively understand the uptake of PFASs in both plant roots and above ground parts, offering key insights into PFAS risk assessment and food safety.

Do Molecular Fingerprints Identify Diverse Active Drugs in Large-Scale Virtual Screening? (No).

Computational approaches for small-molecule drug discovery now regularly scale to the consideration of libraries containing billions of candidate small molecules. One promising approach to increased the speed of evaluating billion-molecule libraries is to develop succinct representations of each molecule that enable the rapid identification of molecules with similar properties. Molecular fingerprints are thought to provide a mechanism for producing such representations. Here, we explore the utility of commonly used fingerprints in the context of predicting similar molecular activity. We show that fingerprint similarity provides little discriminative power between active and inactive molecules for a target protein based on a known active-while they may sometimes provide some enrichment for active molecules in a drug screen, a screened data set will still be dominated by inactive molecules. We also demonstrate that high-similarity actives appear to share a scaffold with the query active, meaning that they could more easily be identified by structural enumeration. Furthermore, even when limited to only active molecules, fingerprint similarity values do not correlate with compound potency. In sum, these results highlight the need for a new wave of molecular representations that will improve the capacity to detect biologically active molecules based on their similarity to other such molecules.

Integrated Transfer Learning and Multitask Learning Strategies to Construct Graph Neural Network Models for Predicting Bioaccumulation Parameters of Chemicals.

Accurate prediction of parameters related to the environmental exposure of chemicals is crucial for the sound management of chemicals. However, the lack of large data sets for training models may result in poor prediction accuracy and robustness. Herein, integrated transfer learning (TL) and multitask learning (MTL) was proposed for constructing a graph neural network (GNN) model (abbreviated as TL-MTL-GNN model) using n-octanol/water partition coefficients as a source domain. The TL-MTL-GNN model was trained to predict three bioaccumulation parameters based on enlarged data sets that cover 2496 compounds with at least one bioaccumulation parameter. Results show that the TL-MTL-GNN model outperformed single-task GNN models with and without the TL, as well as conventional machine learning models trained with molecular descriptors or fingerprints. Applicability domains were characterized by a state-of-the-art structure-activity landscape-based (abbreviated as ADSAL) methodology. The TL-MTL-GNN model coupled with the optimal ADSAL was employed to predict bioaccumulation parameters for around 60,000 chemicals, with more than 13,000 compounds identified as bioaccumulative chemicals. The high predictive accuracy and robustness of the TL-MTL-GNN model demonstrate the feasibility of integrating the TL and MTL strategy in modeling small-sized data sets. The strategy holds significant potential for addressing small data challenges in modeling environmental chemicals.

A New Fingerprint and Graph Hybrid Neural Network for Predicting Molecular Properties.

Machine learning plays a role in accelerating drug discovery, and the design of effective machine learning models is crucial for accurately predicting molecular properties. Characterizing molecules typically involves the use of molecular fingerprints and molecular graphs. These are input into a multilayer perceptron (MLP) and variants of graph neural networks, such as graph attention networks (GATs). Due to the diverse types and large dimension of fingerprints, models may contain many features that are relatively irrelevant or redundant; meanwhile, although the GAT excels in handling heterogeneous graph tasks, it lacks the ability to extract collaborative information from neighboring nodes, which is crucial in scenarios where it cannot capture the joint influence of adjacent groups on atoms. To overcome these challenges, we introduce a hybrid model, combining improved GAT and MLP. In GAT, the recurrent neural network is employed to capture collaborative information. To address the dimensionality issue, we propose a feature selection algorithm, which is based on the principle of maximizing relevance while minimizing redundancy. Through experiments on 13 public data sets and 14 breast cell lines, our model demonstrates superior performance compared to state-of-the-art deep learning and traditional machine learning algorithms. Additionally, a series of ablation experiments were conducted to demonstrate the advantages of our improved version, as well as its antinoise capability and interpretability. These results indicate that our model holds promising prospects for practical applications.