Graphical Features Research Articles

Drug toxicity prediction model based on enhanced graph neural network.

Prediction of drug toxicity remains a significant challenge and an essential process in drug discovery. Traditional machine learning algorithms struggle to capture the full scope of molecular structure features, limiting their effectiveness in toxicity prediction. Graph Neural Network offers a promising solution by effectively extracting drug features from their molecular graphs. However, existing graph learning algorithms fail to account for the interaction features between graph nodes and the indirect edges connecting them. This paper proposes an enhanced graph Neural Network algorithm that employs multi-view features for each node, capturing the feature interactions between each node and its neighbors. Additionally, the adjacency matrix is preprocessed to handle indirect edge interactions. A pooling technique is then applied to aggregate node features, followed by normalization and an activation layer. To further enhance the proposed algorithm, multi-scale attention is applied to learn graph features at different scales, utilizing weights to understand intricate relationships among node feature vectors. The proposed algorithm is evaluated using eight toxicity datasets, covering binary classification, multi-task multi-class, and regression tasks. For binary classification, the Tox21, AMES, Skin reaction, Carcinogens, and DILI datasets are tested. For multi-task multi-class, the ToxCast dataset is applied, and for regression, the LD50 and hREG datasets are tested. The proposed algorithm is compared with four well-known algorithms including Graph Convolution Network, Graph Attention Network, Graph Isomorphism Network, Enhanced Graph Isomorphism Network, and Graph Total Variation. For the classification task, the proposed algorithm achieves ROC-AUC scores of 0.752 for Tox21, 0.775 for AMES, 0.707 for Skin reaction, 0.845 for Carcinogens, 0.92 for DILI, and 0.691 for the ToxCast dataset. For the regression task, the algorithm attains mean square errors of 0.896 for the LD50 dataset and 0.766 for the hREG dataset. These results demonstrate an improvement over the compared algorithms across all evaluated datasets.

Multi-channel spatio-temporal graph attention contrastive network for brain disease diagnosis.

Dynamic brain networks (DBNs) can capture the intricate connections and temporal evolution among brain regions, becoming increasingly crucial in the diagnosis of neurological disorders. However, most existing researches tend to focus on isolated brain network sequence segmented by sliding windows, and they are difficult to effectively uncover the higher-order spatio-temporal topological pattern in DBNs. Meantime, it remains a challenge to utilize the structure connectivity prior in the DBNs analysis. To address these problems, we propose a multi-channel spatio-temporal graph attention contrastive network for DBNs analysis. Specifically, we first construct dynamic brain functional networks from fMRI data with sliding windows, and embed the structural connectivity derived from diffusion tensor imaging (DTI) to the dynamic functional connectivity graph representation to construct multi-modal brain network. Second, we develop a multi-channel spatial attention contrastive network to extract topological features from the brain network within each time window. This network incorporates an intra-window graph contrastive constraint to enhance the discriminative ability of the extracted features. Moreover, temporal dependencies across windows are captured by integrating feature embeddings through a self-attention mechanism, and the inter-window recurrent contrastive constraint is devised to extract higher-order spatio-temporal topological features. Finally, a multi-layer perceptron (MLP) is used to classify the brain networks. Experiments on epilepsy and ADNI datasets show that our method outperforms several state-of-the-art approaches in diagnosing performance, and it provides discriminative graph features for related brain diseases.

Graph Neural Networks for Enhanced Social Network Analysis

Social network analysis (SNA) is an important approach for understanding complex linkages and interactions between entities. Traditional approaches frequently fail to capture the complexities of network data due to its non-Euclidean character. Graph Neural Networks (GNNs) offer an innovative approach to data analysis by modelling node, edge, and graph features using graph structures and neural network topologies. This study investigates the use of GNNs in social network analysis, focusing on problems such as community recognition, impact maximization, link prediction, and sentiment analysis. Our analysis of cutting-edge GNN models shows how they effectively capture and utilize topological and contextual information from social networks

Individualized morphological covariation network aberrance associated with seizure relapse after antiseizure medication withdrawal.

This study intents to detect graphical network features associated with seizure relapse following antiseizure medication (ASM) withdrawal. Twenty-four patients remaining seizure-free (SF-group) and 22 experiencing seizure relapse (SR-group) following ASM withdrawal as well as 46 matched healthy participants (Control) were included. Individualized morphological similarity network was constructed using T1-weighted images, and graphic metrics were compared between groups. Relative to the Control, the SF-group exhibited lower local efficiency, while the SR-group displayed lower global and local efficiency and longer characteristic path length. Both patient groups displayed reduced centrality in certain subcortical and cortical nodes than the Control, with a more pronounced reduction in the SR-group. Additionally, the SR-group exhibited lower centrality of the right pallidum than the SF-group. Decreased subcortical-cortical connectivity was found in both patient groups than the Control, with a more extensive decrease in the SR-group. Furthermore, an edge connecting the right pallidum and left middle temporal gyrus exhibited decreased connectivity in the SR-group than in the SF-group. A weaker small-worldization network upon medication withdrawal, potentially underpinned by node decentralization and subcortical-cortical decoupling, may elevate the risk of seizure relapse.

Overview and Prospects of DNA Sequence Visualization.

Due to advances in big data technology, deep learning, and knowledge engineering, biological sequence visualization has been extensively explored. In the post-genome era, biological sequence visualization enables the visual representation of both structured and unstructured biological sequence data. However, a universal visualization method for all types of sequences has not been reported. Biological sequence data are rapidly expanding exponentially and the acquisition, extraction, fusion, and inference of knowledge from biological sequences are critical supporting technologies for visualization research. These areas are important and require in-depth exploration. This paper elaborates on a comprehensive overview of visualization methods for DNA sequences from four different perspectives-two-dimensional, three-dimensional, four-dimensional, and dynamic visualization approaches-and discusses the strengths and limitations of each method in detail. Furthermore, this paper proposes two potential future research directions for biological sequence visualization in response to the challenges of inefficient graphical feature extraction and knowledge association network generation in existing methods. The first direction is the construction of knowledge graphs for biological sequence big data, and the second direction is the cross-modal visualization of biological sequences using machine learning methods. This review is anticipated to provide valuable insights and contributions to computational biology, bioinformatics, genomic computing, genetic breeding, evolutionary analysis, and other related disciplines in the fields of biology, medicine, chemistry, statistics, and computing. It has an important reference value in biological sequence recommendation systems and knowledge question answering systems.

MMSG-DTA: A Multimodal, Multiscale Model Based on Sequence and Graph Modalities for Drug-Target Affinity Prediction.

Drug-Target Affinity (DTA) prediction is a cornerstone of drug discovery and development, providing critical insights into the intricate interactions between candidate drugs and their biological targets. Despite its importance, existing methodologies often face significant limitations in capturing comprehensive global features from molecular graphs, which are essential for accurately characterizing drug properties. Furthermore, protein feature extraction is predominantly restricted to 1D amino acid sequences, which fail to adequately represent the spatial structures and complex functional regions of proteins. These shortcomings impede the development of models capable of fully elucidating the mechanisms underlying drug-target interactions. To overcome these challenges, we propose a multimodal, multiscale model based on Sequence and Graph Modalities for Drug-Target Affinity (MMSG-DTA) Prediction. The model combines graph neural networks with Transformers to effectively capture both local node-level features and global structural features of molecular graphs. Additionally, a graph-based modality is employed to improve the extraction of protein features from amino acid sequences. To further enhance the model's performance, an attention-based feature fusion module is incorporated to integrate diverse feature types, thereby strengthening its representation capacity and robustness. We evaluated MMSG-DTA on three public benchmark data sets─Davis, KIBA, and Metz─and the experimental results demonstrate that the proposed model outperforms several state-of-the-art methods in DTA prediction. These findings highlight the effectiveness of MMSG-DTA in advancing the accuracy and robustness of drug-target interaction modeling.

Altered Connectome Topology in Newborns at Risk for Cognitive Developmental Delay: A Cross-Etiologic Study.

The human brain connectome is characterized by the duality of highly modular structure and efficient integration, supporting information processing. Newborns with congenital heart disease (CHD), prematurity, or spina bifida aperta (SBA) constitute a population at risk for altered brain development and developmental delay (DD). We hypothesize that, independent of etiology, alterations of connectomic organization reflect neural circuitry impairments in cognitive DD. Our study aim is to address this knowledge gap by using a multi-etiologic neonatal dataset to reveal potential commonalities and distinctions in the structural brain connectome and their associations with DD. We used diffusion tensor imaging of 187 newborns (42 controls, 51 with CHD, 51 with prematurity, and 43 with SBA). Structural weighted connectomes were constructed using constrained spherical deconvolution-based probabilistic tractography and the Edinburgh Neonatal Atlas. Assessment of brain network topology encompassed the analysis of global graph features, network-based statistics, and low-dimensional representation of global and local graph features. The Cognitive Composite Score of the Bayley scales of Infant and Toddler Development 3rd edition was used as outcome measure at corrected 2 years for the preterm born individuals and SBA patients, and at 1 year for the healthy controls and CHD. We detected differences in the connectomic structure of newborns across the four groups after visualizing the connectomes in a two-dimensional space defined by network integration and segregation. Further, analysis of covariance analyses revealed differences in global efficiency (p < 0.0001), modularity (p < 0.0001), mean rich club coefficient (p = 0.017), and small-worldness (p = 0.016) between groups after adjustment for postmenstrual age at scan and gestational age at birth. Moreover, small-worldness was significantly associated with poorer cognitive outcome, specifically in the CHD cohort (r = -0.41, p = 0.005). Our cross-etiologic study identified divergent structural brain connectome profiles linked to deviations from optimal network integration and segregation in newborns at risk for DD. Small-worldness emerges as a key feature, associating with early cognitive outcomes, especially within the CHD cohort, emphasizing small-worldness' crucial role in shaping neurodevelopmental trajectories. Neonatal connectomic alterations associated with DD may serve as a marker identifying newborns at-risk for DD and provide early therapeutic interventions. Trial Registration: ClinicalTrials.gov identifier: NCT00313946.

DeepTGIN: a novel hybrid multimodal approach using transformers and graph isomorphism networks for protein-ligand binding affinity prediction

Predicting protein-ligand binding affinity is essential for understanding protein-ligand interactions and advancing drug discovery. Recent research has demonstrated the advantages of sequence-based models and graph-based models. In this study, we present a novel hybrid multimodal approach, DeepTGIN, which integrates transformers and graph isomorphism networks to predict protein-ligand binding affinity. DeepTGIN is designed to learn sequence and graph features efficiently. The DeepTGIN model comprises three modules: the data representation module, the encoder module, and the prediction module. The transformer encoder learns sequential features from proteins and protein pockets separately, while the graph isomorphism network extracts graph features from the ligands. To evaluate the performance of DeepTGIN, we compared it with state-of-the-art models using the PDBbind 2016 core set and PDBbind 2013 core set. DeepTGIN outperforms these models in terms of R, RMSE, MAE, SD, and CI metrics. Ablation studies further demonstrate the effectiveness of the ligand features and the encoder module. The code is available at: https://github.com/zhc-moushang/DeepTGIN.Scientific contributionDeepTGIN is a novel hybrid multimodal deep learning model for predict protein-ligand binding affinity. The model combines the Transformer encoder to extract sequence features from protein and protein pocket, while integrating graph isomorphism networks to capture features from the ligand. This model addresses the limitations of existing methods in exploring protein pocket and ligand features.

Visual and graphic features of the Russian language edition of Phil O'Neill's book "Poems from the unconscious" as the literature of trauma

The Russian edition project of Phil O'Neill's book "Poems from the Unconscious" is described as a literature of trauma from the perspective of its visual and graphic features. The illustrations and typography of various books related to the literature of trauma are analyzed in order to identify common features in their artistic design and justify the choice of visual and graphic means for the translated edition.

Latin and Cyrillic alphabets in the ergonimicon of the central streets in Kazan: Etymology of names as a reflection of problems and perspectives of graphic representation

This research aims to investigate the graphic features of the central streets’ ergonimicon in Kazan through the application of the linguistic landscape description method, etymological analysis and the method of continuous sampling, which have not been combined in the research of commercial names so far. Their combination is conditioned by the need to confirm the Research Hypothesis: The choice of the alphabet, representing an ergonym, depends on its origin and linguistic affiliation. The relevance of the study is conditioned by the close attention to the linguistic landscapes of the Russian Federation due to the fact that foreign languages and Latin often dominate in this case. The hypothesis and purpose of the study generate the following research questions: RQ1 – Is there an explanation for such an extensive use of Latin script in the linguistic landscapes of the Russian Federation; RQ2 – Is there a pattern in the use of Latin and Cyrillic; RQ3 – Is the prospect of more names in Cyrillic likely to be implemented. The Research Results demonstrated the fruitfulness of combining these methods in ergonimicon research, the hypothesis of the study was confirmed: RR1 – Names belonging to the native lexicon of official languages are transmitted in Cyrillic in the predominant number of cases; RR2 – For ergonyms-internationalisms, the word of the foreign language corresponding in external form and content is almost always chosen; RR3 – Еrgonyms-proper names are written in Latin more often than in Cyrillic; RR4 – Ergonyms-foreign words are represented in Latin. The obtained results led to the following conclusions and opened directions for the research (Conclusions and Directions for Future Research): CDFR1 – the hypothesis of the study should be confirmed on a larger research material; CDFR2 – the search for new approaches to nomination with an emphasis on the native lexicon is recommended; CDFR3 – the need exists for a diachronic approach to the research of linguistic landscapes of the Russian Federation; CDFR4 – the application of programming methods to the description of ergonyms with the aim to solve linguistic problems is required.

Graphical Feature Construction-Based Deep Learning Model for Fatigue Life Prediction of AM Alloys.

Fatigue failure poses a serious challenge for ensuring the operational safety of critical components subjected to cyclic/random loading. In this context, various machine learning (ML) models have been increasingly explored, due to their effectiveness in analyzing the relationship between fatigue life and multiple influencing factors. Nevertheless, existing ML models hinge heavily on numeric features as inputs, which encapsulate limited information on the fatigue failure process of interest. To cure the deficiency, a novel ML model based upon convolutional neural networks is developed, where numeric features are transformed into graphical ones by introducing two information enrichment operations, namely, Shapley Additive Explanations and Pearson correlation coefficient analysis. Additionally, the attention mechanism is introduced to prioritize important regions in the image-based inputs. Extensive validations using experimental results of two laser powder bed fusion-fabricated metals demonstrate that the proposed model possesses better predictive accuracy than conventional ML models.

DFCExpert: Learning Dynamic Functional Connectivity Patterns with Modularity and State Experts.

Characterizing brain dynamic functional connectivity (dFC) patterns from functional Magnetic Resonance Imaging (fMRI) data is of paramount importance in neuroscience and medicine. Recently, many graph neural network (GNN) models, combined with transformers or recurrent neural networks (RNNs), have shown great potential for modeling the dFC patterns. However, these methods face challenges in effectively characterizing the modularity organization of brain networks and capturing varying dFC state patterns. To address these limitations, we propose dFCExpert, a novel method designed to learn robust representations of dFC patterns in fMRI data with modularity experts and state experts. Specifically, the modularity experts optimize multiple experts to characterize the brain modularity organization during graph feature learning process by combining GNN and mixture of experts (MoE), with each expert focusing on brain nodes within the same functional network module. The state experts aggregate temporal dFC features into a set of distinctive connectivity states using a soft prototype clustering method, providing insight into how these states support different brain activities or are differentially affected by brain disorders. Experiments on two large-scale fMRI datasets demonstrate the superiority of our method over existing alternatives. The learned dFC representations not only show improved interpretability but also hold promise for enhancing clinical diagnosis. The code can be accessed at MLDataAnalytics/dFCExpert on GitHub.

Drug–target interactions prediction based on similarity graph features extraction and deep learning

Drug–target interactions prediction based on similarity graph features extraction and deep learning

Spectrum‐sensing algorithm based on graph feature fusion

AbstractGraph‐based spectrum sensing in noisy environments has major implications for civilian and military signal processing applications. However, existing algorithms suffer from high computational complexity and performance deterioration at low signal‐to‐noise ratios (SNRs). Therefore, a spectrum‐sensing algorithm based on graph feature fusion using a quadratic form derived from self‐loop weights and the graph Laplacian matrix is proposed in this study. The sum of the first and second block maxima of the power spectrum of the observed signal is selected as the input to the graph converter. Self‐loop weights are combined with the Laplacian matrix to construct the graph quadratic form, which serves as the test statistic for decision‐making. By applying majorisation and the extreme value theory, it is demonstrated that the proposed algorithm outperforms existing methods. The simulation results confirm the robust spectrum‐sensing performance across various signal modulation types and pulse shapes. Thus, compared to existing algorithms, except block range‐ and energy‐detection‐based methods, the proposed algorithm demonstrates the best spectrum‐sensing performance under low SNRs and channel‐fading conditions, while achieving the lowest computational complexity. The proposed approach enables more efficient and accurate spectrum sensing, fostering advancements in communication technologies and defence applications.

Design of Traditional Inkblot Pattern Plane System Based on Graphics and Visual Perception

In order to improve the graphic design effect of traditional pattern with ink and wash elements, this paper puts forward a graphic and visual perception-based graphic system design scheme of traditional pattern with ink and wash elements. Taking plane space media parameters, creative display patterns, characters and other elements as constraint parameters, this paper constructs a plane design model under traditional pattern fusion ink elements, extracts the graphic and visual perception element feature space structure of plane design graphics under traditional pattern fusion ink elements, and combines the three-dimensional internal structure characteristics of plane design graphics under traditional pattern fusion ink elements to carry out information reorganization. The method of color space color feature decomposition is used to decompose the visual elements of graphic design under traditional pattern fusion ink elements, and the virtual reconstruction model of graphic design under traditional pattern fusion ink elements and RGB-D image of visual perception element is established by the method of image aesthetic and emotional factors fusion. The feature recombination method of converting graphic design into materialized image is used to design the three-dimensional reconstruction model of graphic design under traditional pattern fusion ink elements. The 3D multi-viewpoint tracking and recognition method is adopted to reconstruct the graphic design and visual perception elements under the traditional pattern fusion ink elements, so as to improve the graphic design ability under the traditional pattern fusion ink elements. The simulation results show that this method has good visual expression ability, good graphic design effect, and good visual communication ability of graphic design with traditional pattern and ink elements.

Unified representation aggregated from graph and global features for multi-source data clustering

Multi-source data has received increasing attention due to its excellent performance in clustering tasks. However, existing multi-source data clustering methods utilize shallow graph learning methods to model similarity graphs of multi-source data, and ignore the importance of weak connections between samples within the same cluster. Thus, these connections fail to be explored in the global similarity graph of multi-source data. In this paper, we proposed a unified representation aggregated from graph and global features (UGGF) method for multi-source data clustering. Specifically, it utilizes the cross-source graph diffusion process to preserve invariant connections in multi-source similarity graphs and preserve weak connections between samples through their connection relationships between different sources. Furthermore, inspired by self-attention, the encoder of transformers is used to learn multi-source global feature representation. Then, global features and similarity graphs are integrated to comprehensively explore the multi-source data, aiming to obtain a unified representation for the final clustering task. The proposed UGGF method is validated on four benchmark datasets. Extensive experimental results demonstrate the superiority of the proposed method compared with other state-of-the-art multi-source data clustering methods.

Propofol‐induced moderate–deep sedation modulates pediatric neural activity: A functional connectivity study

AbstractBackgroundPrevious studies have demonstrated the underlying neurophysiologic mechanism during general anesthesia in adults. However, the mechanism of propofol‐induced moderate–deep sedation (PMDS) in modulating pediatric neural activity remains unknown, which therefore was investigated in the present study based on functional magnetic resonance imaging (fMRI).MethodsA total of 41 children (5.10 ± 1.14 years, male/female 21/20) with fMRI were employed to construct the functional connectivity network (FCN). The network communication, graph‐theoretic properties, and network hub identification were statistically analyzed (t test and Bonferroni correction) between sedation (21 children) and awake (20 children) groups. All involved analyses were established on the whole‐brain FCN and seven sub‐networks, which included the default mode network (DMN), dorsal attentional network (DAN), salience network (SAN), auditory network (AUD), visual network (VIS), subcortical network (SUB), and other networks (Other).ResultsUnder PMDS, significant decreases in network communication were observed between SUB‐VIS, SUB‐DAN, and VIS‐DAN, and between brain regions from the temporal lobe, limbic system, and subcortical tissues. However, no significant decrease in thalamus‐related communication was observed. Most graph‐theoretic properties were significantly decreased in the sedation group, and all graphical features of the DMN showed significant group differences. The superior parietal cortex with different neurological functions was identified as a network hub that was not greatly affected.ConclusionsAlthough the children had a depressed level of neural activity under PMDS, the crucial thalamus‐related communication was maintained, and the network hub superior parietal cortex stayed active, which highlighted clinical practices that the human body under PMDS is still perceptible to external stimuli and can be awakened by sound or touch.

Discovery of novel VEGFR2 inhibitors against non-small cell lung cancer based on fingerprint-enhanced graph attention convolutional network

Despite the proven inhibitory effects of drugs targeting vascular endothelial growth factor receptor 2 (VEGFR2) on solid tumors, including non-small cell lung cancer (NSCLC), the development of anti-NSCLC drugs solely targeting VEGFR2 still faces risks such as off-target effects and limited efficacy. This study aims to develop a novel fingerprint-enhanced graph attention convolutional network (FnGATGCN) model for predicting the activity of anti-NSCLC drugs. Employing a multimodal fusion strategy, the model integrates a feature extraction layer that comprises molecular graph feature extraction and molecular fingerprint feature extraction. The performance evaluation results indicate that the model exhibits high accuracy and stability in predicting activity. Moreover, we explored the relationship between molecular features and biological activity through visualization analysis, thus improving the interpretability of the approach. Utilizing this model, we screened the ZINC database and conducted high-precision molecular docking, leading to the identification of 11 potential active molecules. Subsequently, molecular dynamics simulations and free energy calculations were performed. The results demonstrate that all 11 aforementioned molecules can stably bind to VEGFR2 under dynamic conditions. Among the short-listed compounds, the top six exhibited satisfactory inhibitory activity against VEGFR2 and A549 cells. Especially, compound Z-3 displayed VEGFR2 inhibitory with IC50 values of 0.88 μM, and anti-proliferative activity against A549 cells with IC50 values of 4.23 ± 0.45 μM. This approach combines the advantages of target-based and phenotype-based screening, facilitating the rapid and efficient identification of candidate compounds with dual activity against VEGFR2 and A549 cell lines. It provides new insights and methods for the development of anti-NSCLC drugs. Furthermore, further biological activity tests revealed that Z1-Z3 and Z6 manifested relatively strong antiproliferative activities against NCI-H23 and NCI-H460, and relatively low toxicity towards GES-1. The hit compounds were promising candidates for the further development of novel VEGFR2 inhibitors against NSCLC.

Iterative Deep Structural Graph Contrast Clustering for Multiview Raw Data.

Multiview clustering has attracted increasing attention to automatically divide instances into various groups without manual annotations. Traditional shadow methods discover the internal structure of data, while deep multiview clustering (DMVC) utilizes neural networks with clustering-friendly data embeddings. Although both of them achieve impressive performance in practical applications, we find that the former heavily relies on the quality of raw features, while the latter ignores the structure information of data. To address the above issue, we propose a novel method termed iterative deep structural graph contrast clustering (IDSGCC) for multiview raw data consisting of topology learning (TL), representation learning (RL), and graph structure contrastive learning to achieve better performance. The TL module aims to obtain a structured global graph with constraint structural information and then guides the RL to preserve the structural information. In the RL module, graph convolutional network (GCN) takes the global structural graph and raw features as inputs to aggregate the samples of the same cluster and keep the samples of different clusters away. Unlike previous methods performing contrastive learning at the representation level of the samples, in the graph contrastive learning module, we conduct contrastive learning at the graph structure level by imposing a regularization term on the similarity matrix. The credible neighbors of the samples are constructed as positive pairs through the credible graph, and other samples are constructed as negative pairs. The three modules promote each other and finally obtain clustering-friendly embedding. Also, we set up an iterative update mechanism to update the topology to obtain a more credible topology. Impressive clustering results are obtained through the iterative mechanism. Comparative experiments on eight multiview datasets show that our model outperforms the state-of-the-art traditional and deep clustering competitors.

The interaction of technological affordances and user preferences: A corpus-based study of graphic features across Twitter and Discord

This paper investigates the influence of communication device on linguistic variation across two levels of scale. Drawing on the notion of affordances (Hutchby, 2001), it explores how technological differences between the computer and smartphone may shape, without ultimately determining, the use of graphic features. To do so, the empirical study uses the methodological approach of ‘scaling down’ the analysis of two features across device types: first letter capitalisation and emoji. The first step of analysis is purely quantitative, focusing on linguistic variation in a corpus of a million messages from Twitter (now X). The second step of analysis uses quantitative and qualitative methods to investigate the same features within several thousand messages, by two known participants, across the platforms Twitter and Discord. Thus, the study integrates initial results from a large, anonymous corpus with a much smaller but more detailed one. Overall, device affordances are found to influence feature use; however, statistical tendencies regarding variation across device type do not necessarily hold true for each dimension of context or individual. The combination of different scales thus leads to a more thorough understanding of the influence of technology on linguistic variationand its limits in favour of contextual factors and individual preferences.