Browse Articles

Processing-dependent reinforcing behavior of garlic peel and pineapple leaf fibers in poly(butylene succinate) all-biocomposites

The effects of continuous chest compression protocol on cardiopulmonary resuscitation (CPR) quality and fatigue

Aim This study investigated how fatigue (measured both subjectively and objectively) and cardiopulmonary resuscitation (CPR) quality are impacted by various CPR protocols, principally comparing continuous chest compressions with intermittent rest periods. Methods A sample of 23 participants (M = 22.3 years ± 3.8) was used in a cross-sectional design. Four compression:rest ratios were analyzed: 30:10, 50:10, 100:10, and continuous compressions (CC), with all participants performing CPR for 9 minutes in each condition. Subjective fatigue was measured using the Borg Rating of Perceived Exertion (RPE) at baseline, three, six, and nine minutes. Objective fatigue data were collected using the Parvo metabolic cart and heart rate (HR) monitors. Regression models of CPR performance quality (i.e., depth, recoil, and rate) investigated the effect of CPR protocol while controlling for participant variables. Fatigue (i.e., RPE, HR, and VO 2 ) were compared across CPR protocols using repeated measure one-way ANOVA and post hoc Tukey’s honestly significant difference (HSD) tests. Results Significant differences in performance were observed between CC and other protocols: 30 (p < .001), 50 (p = .009), and 100 (p = .031). Percent of full chest recoil was also significantly lower during CC than in 30, 50, and 100 conditions (all p < .001). Subjective fatigue increased significantly with longer compression ratios (p < .001). Conclusion Higher compression ratios result in greater perceived fatigue and reduced CPR performance quality. Rescuer fatigue should be considered in CPR protocol design to optimize both rescuer effectiveness and patient outcomes.

Vector OBF detection method of magnetic tensor for small magnetic targets

Abstract To improve the detection probability of small magnetic targets and enrich the direction information in magnetic scalar orthonormal basis function (OBF) detection, this paper proposes a vector OBF detection method using the magnetic tensor. Firstly, the OBF expression of the magnetic tensor is determined by analyzing the OBF decomposition principle of the magnetic tensor vector. Then, the magnetic tensor vector OBF detection model is constructed, and the magnetic tensor vector OBF detection method is proposed. Finally, the AR whitening filter is investigated to optimize the proposed method. The detection method can detect small magnetic targets under the condition of a low signal-to-noise ratio (SNR). The simulation results indicate that the detection accuracy of the scalar OBF detection method is significantly decreased under the Gaussian noise with an SNR of −19 dB, while the vector OBF detection method can still effectively detect magnetic anomaly signals; when detecting the same target under the false alarm probability of 1.5% and different fractal noises, the detection probability of the vector OBF detection method is 6 ~ 35% higher than that of the scalar OBF detection method.

Tool choice matters: Evaluating edgeR vs. DESeq2 for sensitivity, robustness, and cross-study performance

Differential gene expression (DGE) analysis is foundational to transcriptomic research, yet tool selection can substantially influence results. This study compares two widely used DGE tools, edgeR and DESeq2 , using real and semi-simulated bulk RNA-Seq data sets, mostly from human patients, spanning viral infection, bacterial infection, and fibrotic conditions. We evaluated tool performance across four dimensions: (1) sensitivity to sample size and robustness to outliers; (2) classification performance of uniquely identified gene sets within the discovery dataset; (3) pathway-level concordance of significant DEG sets; and (4) generalizability of tool-specific gene sets across independent studies. First, using Bonferroni-adjusted p -value < 0.05 and absolute log 2 fold change greater than 1 (i.e., | log 2 FC | > 1 ) as significance criteria, repeated subsampling showed that DESeq2 generally identified more Differentially Expressed Genes (DEGs) than edgeR at smaller sample sizes, while the tools became more concordant as sample size increased. Both tools showed similar responses to simulated outliers, with Jaccard similarity decreasing as more swapped samples were introduced. Second, classification models trained on tool-specific genes showed that edgeR achieved higher F1 scores in 9 of 13 contrasts and more frequently reached perfect or near-perfect precision. Third, Hallmark and KEGG pathway enrichment analyses showed that many contrasts retained substantial pathway-level agreement between tools, although selected contrasts still showed tool-specific enriched pathways. Finally, in cross-study validation using four independent SARS-CoV-2 datasets, edgeR -specific genes yielded higher AUC, precision, and recall in held-out datasets, with some test cases achieving perfect separation. Overall, our findings show that DESeq2 may identify more DEGs under stringent thresholds, whereas edgeR often yields more conservative, predictive, and generalizable gene sets. These findings emphasize that DGE tool choice should be guided not only by DEG yield, but also by the downstream reproducibility, predictive value, and biological interpretability of the resulting gene sets.

A lightweight explainable deep learning framework for coal classification

Abstract Accurate and efficient coal classification is essential for optimizing energy production, improving resource utilization, and ensuring environmental compliance in industrial applications. Traditional methods that rely on visual inspection and laboratory analysis are often slow, subjective, and difficult to scale. In response to these challenges, this research proposes a custom task adaptive lightweight convolutional neural network that integrates Separable Convolutional Blocks, Inverted Residual Blocks, and Squeeze and Excitation mechanisms to achieve high classification performance with low computational overhead. The model is trained and evaluated on the large scale DsCGF dataset, which comprises over 270,000 images of coal, gangue, foreign objects, and unknown materials collected under real world production and non-production conditions from multiple mining regions. The experimental results show that the proposed model achieves classification accuracies of 93.73% on Anhui Guobei production dataset, 99.97% on the Anhui Guobei non-production dataset, 99.26% on the Inner Mongolia Erlintu production dataset, and 90.50% on the challenging Shanxi Wangjialing production dataset, while using only 1.72 million parameters and achieving an average inference time of 0.023 s per image. Compared with well-established transfer learning models such as MobileNetV2, DenseNet201, and Xception, the proposed architecture consistently demonstrates superior or comparable performance across accuracy, precision, recall, and F1 score metrics with significantly reduced computational complexity. To enhance interpretability and provide visual insights into model predictions, explainable artificial intelligence techniques including Grad-CAM, Grad-CAM++, and Score-CAM are employed to visualize class discriminative regions. The proposed framework provides a practical, interpretable, and resource efficient solution for coal classification and contributes meaningfully to the advancement of intelligent mining systems.

A dual-stream deep learning architecture for business impact scoring and alert escalation

Modern network monitoring systems generate massive volumes of telemetry data, yet most existing anomaly detection models fail to prioritize alerts according to their operational urgency and business impact. This limitation results in delayed incident responses and inefficient alert management in Network Operations Centers. To address this gap, this study proposes a Dual-Stream Predictive Alert Escalation Framework that integrates temporal failure pattern learning with business impact-aware alert prioritization. The proposed architecture consists of two key components: a bidirectional temporal encoder for modeling multivariate Key Performance Indicator (KPI) time-series data, and an auxiliary severity encoder that captures contextual metadata related to operational risk and service criticality. The outputs of these two learning streams are combined through an attention-based fusion mechanism, and a Business Impact Scoring (BIS) layer generates impact-weighted escalation decisions for proactive incident management. Experimental evaluations using real-world KPI datasets and the AI4I_2020 predictive maintenance dataset demonstrate the superior performance of the proposed framework compared to baseline methods such as LSTM, GRU, CNN-LSTM, and BiLSTM-VAE. On the combined multivariate KPI dataset, the model achieved a precision of 0.94, a recall of 0.92, and an F1-score of 0.93, along with a PR-AUC of 0.95 and a ROC-AUC of 0.94. Under impact-aware evaluation, the framework attained the highest Impact-Weighted F1 (IW-F1) of 0.85 and BIS accuracy of 0.88, resulting in an estimated 31.6% reduction in operational costs through earlier and more accurate escalation of critical events. The suitability of the selected datasets is justified by their complementary roles: publicly available KPI time-series datasets represent real-world network telemetry behavior, while the AI4I_2020 dataset provides structured severity and operational context, enabling joint evaluation of failure prediction accuracy, escalation timeliness, and business impact modeling. By prioritizing alerts based on impact-aware severity rather than raw anomaly scores, the proposed framework directly supports operational cost reduction through earlier mitigation and improved decision-making in network operations. The proposed approach bridges the gap between anomaly detection and intelligent alert management by incorporating business relevance into predictive modeling. This dual-stream architecture offers a scalable and proactive solution for AIOps-driven network reliability and automated service resilience.

A sex hormone-BDNF-TrkB axis directs sympathetic innervation in the mammary gland

Structural equation modeling for exploring the barriers to accessing healthcare among hijra in Bangladesh

Introduction Hijra in Bangladesh and other South Asian countries face considerable barriers to healthcare given their potential exposure to repeated, multi-level stigmatization and marginalization. However, these needs remain unaddressed in the healthcare system, thus warranting exploration of their healthcare access barriers. This article examines healthcare access barriers among hijra in Bangladesh. Methods A cross-sectional survey was conducted from June to November 2021 among 544 hijra across 16 districts out of 64 districts (25%) across all eight divisions in Bangladesh using multi-stage stratified cluster sampling, thus spanning a wide geographical coverage. Among all the participants, those who reported health problems in the last six months and sought treatment as a hijra (N = 215) were further used for analysis. Structural equation modeling (SEM) was utilized to assess and quantify the pathways how harassment, financial difficulties, discrimination, fear of discrimination, and lack of infrastructural facilities significantly creates barriers in accessing healthcare facilities among hijra . The results were expressed using factor loadings (FL) where FL closer to 1 indicated strong relationship. Results The average age of hijra participants was 32 years old (±10.3 SD). The participants reported an average monthly income of US$107.2. Findings revealed that 53.5% of hijra respondents encountered four types of healthcare access barriers: harassment, discrimination, financial difficulties, and lack of infrastructural facilities in some form. Among these healthcare access barriers, our model illustrated that three were significant. These were discrimination by healthcare providers (Factor Loading = 1.14), which was the most significant barrier, followed by harassment by healthcare providers (Factor Loading = 0.93) and lack of infrastructural facilities (Factor loading = 0.41), and all these three barriers lead to an ultimate barrier to access to healthcare for the hijra community. Conclusion This study highlights the critical barriers faced by hijra in accessing healthcare in Bangladesh. Findings suggest that interventions aimed at improving healthcare access should simultaneously address multiple factors to ensure gender-responsive healthcare systems that do not conform to a rigid gender binary.

Temporal context shapes multicue integration in auditory time-to-contact judgments

Abstract Anticipating when a moving object will arrive is fundamental for action. While vision-based time-to-contact (TTC) judgment follows reliability-weighted integration of motion and duration cues, it remains unclear whether audition supports similar computations. We asked listeners to judge TTC for moving sounds in motion-only, duration-only, and combined cue conditions, under fast and slow temporal contexts. Combined cues consistently improved precision, but temporal context reshaped bias. Indeed, integration generates remarkable delays in the fast regime and reduced anticipatory errors in the slow regime. Regression-to-the-mean effects revealed strong influences of context-dependent priors. A hierarchical model further suggested that adjusted cue weights in a manner broadly consistent with reliability-based cue use, while deviations from the flat-prior benchmark indicated that temporal context and priors shaped performance. These results indicate that auditory TTC is not a fixed computation, but an adaptive inference process shaped by temporal context, revealing a modality-specific route to predictive timing.

Mapping global inequities in telemedicine implementation: An umbrella review of barriers and facilitators

Telemedicine expanded rapidly during the COVID-19 pandemic, yet its diffusion has remained uneven across health systems. Whether implementation challenges differ systematically across economic contexts, and what implications these differences hold for global digital health equity, has not been comprehensively characterized. This study presents an umbrella review (PROSPERO CRD42024615998) of systematic reviews and meta-analyses indexed in PubMed and Scopus through December 2025 that reported barriers and/or facilitators to telemedicine implementation. Methodological quality was appraised using R-AMSTAR. Reported items were extracted verbatim, embedded with the Universal Sentence Encoder, and grouped into thematic clusters using HDBSCAN density-based clustering, with parameters optimized by SLSQP for cluster cohesion (median intra-cluster similarity ≥ 0.5). Analyses were stratified by World Bank country income group to characterize context-specific implementation profiles, identify evidence gaps, and detect orphaned barriers, defined as challenges lacking any corresponding documented facilitator. A total of 161 systematic reviews were included, yielding 1,333 barriers and 504 facilitators (corresponding to a barrier-to-facilitator ratio of 2.6:1). Across all settings, the most frequently reported barriers were high costs (n = 106), technical issues (n = 94), and training and knowledge (n = 91). Thematic profiles differed markedly by income group: high-income countries reported predominantly second-generation challenges centered on workflow integration, interoperability, and user experience, whereas lower-middle-income countries reported foundational barriers centered on infrastructures and costs. Evidence produced from low-income countries was entirely absent. Several high-frequency barriers lacked corresponding facilitator clusters, indicating challenges for which the published literature provides no documented solutions. The global telemedicine evidence base is itself inequitably distributed and risks reinforcing the disparities it is intended to reduce. High-income settings require implementation research focused on human and organizational factors, whereas lower-income settings require foundational research and investment in basic infrastructures before workflow-level optimization becomes meaningful. Without context-specific, equity-oriented strategies, telemedicine risks widening rather than narrowing the global digital health divide.

The ability of Lactiplantibacillus plantarum PK 1.1 to synthesize odd-chain and cyclic fatty acids in oat-based beverages

Abstract This study aims to optimize the composition of a model oat beverage to enhance the production of minor fatty acids, including odd-chain and cyclic fatty acids. The oat beverage composition was optimized in the Box-Behnken design, varying by the ratio of oat protein hydrolysate (0, 1, 2%), coconut oil (0, 1.5, 3.0%), and added sucrose (0, 2, 4%), while maintaining a constant ratio of oat β-glucan (2%) and oat protein (2%). In the first stage of the study, fermentation was conducted at 22 °C for 48 h using the Lactiplantibacillus plantarum PK 1.1 strain. The four best variants that promoted the biosynthesis of rare fatty acids were tested in a refined experiment. In this stage, the effects of low-temperature stress (15 °C) and osmotic stress induced by the addition of NaCl (up to 5%) were assessed. The prepared formulations were analysed for lactic acid bacteria (LAB) counts, titratable acidity, pH, apparent viscosity, and fatty acid composition. In the first experiment, the highest combined share of odd-chain and cyclic fatty acids (up to 0.353%) was observed in beverages prepared without coconut oil. This low concentration prompted a second experimental stage in which microbial stress was intensified by fermentation at 15 °C and by osmotic stress induced through NaCl addition (0%, 2.5%, or 5%). In this stage, the total share of these fatty acids increased to 1.51%, with cyclic fatty acids predominating. In both experiments LAB counts, titratable acidity, pH, and apparent viscosity were comparable to those of commercially available fermented plant-based beverages. We recommend further research to elucidate the metabolic and technological factors that influence the odd-chain and cyclic fatty acids ratio under suboptimal fermentation temperatures.

Genome-wide characterization of copy number variants and their functional relevance in indigenous draught cattle of South Asia

Copy number variations (CNVs) are an important source of structural genomic variation and contribute to phenotypic diversity in cattle, including traits related to production, reproduction, and adaptation. In this study, we performed a genome-wide characterization of copy number variation regions (CNVRs) in Asian zebu cattle representing eight indigenous draught breeds traditionally used for ploughing, wet-field agriculture, and carting. CNVs were detected using two read-depth–based approaches, CNVnator and CNVcaller, which identified 7,705 and 5,640 CNVs, respectively. Integration of results from both methods using a 50% reciprocal overlap criterion yielded 6,143 CNVRs. The average number of CNVRs per breed ranged from 4,607–5,005, collectively covering approximately 6.23% of the autosomal genome. Of these, 2,697 CNVRs were shared across all breeds, whereas 190 CNVRs were breed specific. Population differentiation based on CNVs, estimated using pairwise V ST statistics, indicated that the Hallikar breed exhibited the highest average differentiation (0.07) relative to other breeds. A total of 4,868 genes overlapped with the identified CNVRs and were enriched for biological processes associated with immune regulation, metabolic function, and adaptive responses. Further comparison with cattle quantitative trait loci identified 65 unique QTLs, predominantly linked to carcass, fertility, reproduction, and growth traits. Overall, this study describes the genome-wide distribution and diversity of CNVRs in South Asian indigenous draught cattle and provides baseline genomic information for further investigations into structural variations relevant to breeding, management, and conservation.

High-performance liquid chromatography – diode array detection method validation for amentoflavone-type biflavonoids in five Encephalartos species with potential neuroprotective activity

Abstract Ginkgo biloba is a well-known food supplement for enhancing memory and is rich in biflavonoids. Biflavonoids are predominant in the Order Cycadales (cycads). Among cycads, five Encephalartos species, E. ferox , E. kisambo , E. laurentianus , E. natalensis , and E. villosus , were selected to assess the neuroprotective potential by estimation of the antioxidant and acetylcholinesterase (AChE) inhibition activities. E. natalensis and E. ferox strongly inhibited AChE (IC 50 =1.349 ± 0.041 and 1.948 ± 0.06 µg/mL respectively), while E. kisambo and E. ferox were the most potent antioxidant. Accordingly, E. ferox was selected for further investigation. Chromatographic isolation of the ethyl acetate fraction afforded amentoflavone ( 1 ), bilobetin ( 2 ), ginkgetin ( 3 ), naringenin ( 6 ), and apigenin( 7 ). The isolated biflavonoids ( 1–3 ) showed potent AChE inhibitory activity with IC 50 = 2.146 ± 0.086, 0.762 ± 0.039, and 1.474 ± 0.061 µg/mL respectively compared to rivastigmine, the positive control (IC 50 = 3.357 ± 0.103 µg/mL). A validated high-performance liquid chromatography with diode array detection (HPLC-DAD) method was developed for the simultaneous estimation of five biflavonoids: amentoflavone, bilobetin, ginkgetin, isoginkgetin and sciadopitysin in the five Encephalartos species and Ginkgo biloba. In addition, total phenolic and total flavonoid contents were estimated in the selected plants. This study highlights Encephalartos as a potential source of bioactive compounds for further neuroactive drug discovery research.

Assessment of simulation-based inference methods for stochastic compartmental models in epidemiological research

Global pandemics, such as the recent COVID-19 crisis, highlight the need for stochastic epidemic models that can capture the randomness inherent in the spread of disease. Such models must be accompanied by methods for estimating parameters in order to generate fast nowcasts and short-term forecasts that can inform public health decisions. This paper presents a comparison of two advanced Bayesian inference methods: 1) pseudo-marginal particle Markov chain Monte Carlo, using an unbiased likelihood estimate obtained by Particle Filter (PF), and 2) Conditional Normalizing Flows (CNF). We investigate their performance on three commonly used compartmental models: A classical Susceptible-Infected-Susceptible (SIS), a Susceptible-Infected-Recovered (SIR) model and a two-variant Susceptible-Exposed-Infected-Recovered (SEIR) model, complemented by an observation model that maps latent trajectories to empirical data. Addressing the challenges of intractable likelihoods for parameter inference in stochastic settings, our analysis highlights how particle-filter-based likelihood estimation and flow-based posterior approximation can provide accurate and robust inference capabilities. The results of our simulation study further underscore the effectiveness of these approaches in capturing the stochastic dynamics of epidemics, providing prediction capabilities for the control of epidemic outbreaks. Results on an Ethiopian cohort study demonstrate operational robustness under real-world noise and irregular data sampling. To facilitate reuse and to enable building pipelines that ultimately contribute to better informed decision making in public health, we make code and synthetic datasets publicly available.

Pedestrian mobility citizen science complements expert mapping for enhancing inclusive neighborhood

Effects of walking training with blood flow restriction on the hemodynamics and perceptual responses among sedentary college students: A randomized crossover trial

College students often present a sedentary lifestyle. Low-intensity walking training with blood flow restriction (WT-BFR) may offer health benefits comparable to moderate-to-high intensity walking training without restriction, yet its effects in sedentary college students remain unclear. This study aimed to examine the effects of WT-BFR at different limb occlusion pressure (LOP) on hemodynamic and perceptual responses in sedentary college students using a randomized crossover design. The study was registered on the China Clinical Trial Registry (ChiCTR2500097728 25/02/2025). A total of 60 participants completed the 5-minute WT-BFR with varying LOPs (i.e., 0%, 40%, 60%, and 80%). Hemodynamic parameters (blood pressure and heart rate) were measured before, immediately after, and 5 minutes post-intervention. Meanwhile, perceptual responses (perceived exertion and discomfort) and step numbers were recorded post-intervention. For the hemodynamic parameters, only 60% LOP showed a larger increase in heart rate after training than 0% LOP (walk training without BFR) condition (3.82, 95%CI: 0.72 to 6.91, p = 0.016, beats/min), representing a relative increase of approximately 5.6% from baseline. With the increase of LOP, perceived exertion and discomfort were increased significantly (p < 0.05), and the step numbers were reduced (p < 0.05). Based on participant perception, an LOP range of 40%–60% is recommended. Clinically, the observed cardiovascular changes were modest and within safe ranges for healthy young adults. The long-term health effects of low-intensity WT-BFR among college students warrant further investigation.

Prevalence of celiac disease among permanent residents in Northwest China: A cross-sectional survey

Abstract Celiac disease (CeD) is an autoimmune small bowel disease with prevalence varying by region and population. There is a lack of large-scale epidemiological data on CeD among permanent residents in Northwest China. This study aimed to explore the demographic, epidemiological, and laboratory features of CeD in this region, to support its prevention and control. From June 2022 to December 2023, a cross-sectional survey on CeD prevalence was carried out among 4602 permanent residents in 12 administrative districts of Northwest China. Serum anti-tissue transglutaminase IgA antibody levels were detected by chemiluminescence. For antibody-positive patients, gastroscopic duodenal histopathological examination was performed for confirmatiom of diagnosis. Demographic date, laboratory data, comorbidities, and medical history were collected and statistically analyzed. The serum prevalence of CeD was 1.26% (58/4602, 95% CI 0.96%–1.62%), while the biopsy-diagnosed prevalence was 1.02% (47/4602, 95% CI 0.75%–1.36%). Females had a higher prevalence than males (1.39% vs. 0.44%, P  = 0.002). Among ethnic groups, Kazakhs had the highest CeD prevalence (2.74%), followed by Tajiks, Uygurs, and Han Chinese ( P  < 0.001). Those at latitudes 40°N or higher were more likely to have CeD than those below 40°N ( P  = 0.001). In Xinjiang, North Xinjiang had the highest prevalence (1.66%), followed by East and South Xinjiang ( P  < 0.001). There were no significant differences in prevalence by age, BMI, or education. CeD prevalence was higher in the nephropathy group (6.67%). In Northwest China, the biopsy-diagnosed CeD prevalence was 1.02%. Prevalence differed by sex, ethnicity, latitude, and location. There was a significant link between CeD and renal diseases. Serological screening in high-prevalence areas is essential for early CeD diagnosis.

Strategies for implementing genomic selection in a public soybean breeding program

Improving selection accuracy in soybean breeding programs is crucial for reducing costs and shortening the time required to develop new varieties. Genomic selection (GS) is a promising tool for enhancing the accuracy of selection. This study aimed to identify the most effective GS models for predicting key traits, including seed yield, protein content, oil content, and maturity in soybean breeding programs. Additionally, we sought to determine the optimal training population optimization method (structure vs. size) and establish the minimum number of genotypes required for a training population that ensures high model performance across locations. Finally, we explored multitrait selection based on genomic prediction to improve breeding decisions. Data were obtained from the soybean variety development program and included experiments planted in a randomized block design with two replications at eight locations during the 2023 and 2024 growing seasons. Six GS models were tested: rrBLUP, Bayes A, Bayes B, RKHS, random forest, and support vector machine. Four methods for optimizing the training population were evaluated: random selection (RS), maturity group random selection (MGRS), experimental random selection (ERS), and genetic algorithm (GA). We also assessed ten different training population sizes (ranging from 10% to 90%) and five selection index strategies: direct selection on seed yield, direct selection on oil content, direct selection on protein content, Rank-index, and Smith-Hazel index. Our results suggest that the most effective strategy for implementing GS in a public soybean breeding program involves using the rrBLUP model for genomic prediction. To optimize its performance, it is recommended to train the model using 80% of the total population. This approach provides a robust and reliable prediction. Furthermore, structuring the training population through experimental random selection enhances genetic diversity, which is crucial for improving selection accuracy and robustness across breeding cycles. Finally, the Rank-index proved to be a highly effective strategy for selecting soybean genotypes, particularly for improving seed yield and oil content, while results for seed protein content were less promising. By considering multiple traits simultaneously, this method offers a more balanced approach to genetic improvement compared to single-trait selection methods, making it an excellent tool for breeding programs that aim to enhance both productivity and quality.

Insights into the toxic effects of moxifloxacin in Allium cepa through multiparametric experimental and in silico analyses

T-pGNN4DTI: Towards better drug-target interactions prediction using Global Self-attentive Pooled Graph Convolutional Networks and protein pre-training Models

Identification of drug-target interactions (DTI) is an important and challenging task in drug discovery and development. Traditional methods generally require biological experiments, which are costly and time-consuming. Machine learning-based methods can rapidly predict DTI using only computer algorithmic models, allowing researchers to validate only the most promising interactions through biochemical experiments. This holds promise for effectively addressing the current challenges of lengthy development cycles and high costs in new drug development. However, it is difficult for the existing DTI prediction methods to learn complete and effective feature information from the compound and protein. Therefore, this work proposes a DTI prediction method based on the global self-attentive pooled graph neural network and protein pretraining model, called T-pGNN4DTI. On the one hand, T-pGNN4DTI uses a global self-attention pooled graph neural network to learn more meaningful features of the drug molecule by paying more attention to the information features of certain important atomic nodes of the molecular structure and ignoring some weakly relevant node information features. On the other hand, T-pGNN4DTI uses a pre-trained Transformer-based model to capture the semantic relationships of contexts in long sequences of proteins, which can learn more complete feature information. The results of comparing experiments on three benchmark datasets show that the performance of the proposed T-pGNN4DTI model is better than that of the existing DTI prediction methods, effectively improving the DTI prediction. It provides a new way of thinking to help solve the DTI-related problems.