A team from Chengdu University and collaborating hospitals developed a gradient boosting machine learning model to assess sleep disorder risk in older adults with multimorbidity. By integrating demographic, clinical, and behavioral data, and using SHAP values for interpretability, the model highlights pivotal predictors such as frailty, cognitive function, and nutritional status to support targeted interventions.

Key points

• Applied gradient boosting machine on 471 multimorbid seniors, achieving AUC=0.881 for sleep disorder risk prediction.
• Employed LASSO and Boruta for feature selection, identifying seven predictors: frailty, cognitive status, nutritional status, living alone, depression, smoking, and anxiety.
• Used SHAP analysis for model interpretability, quantifying each feature’s contribution to facilitate personalized risk assessment.

Why it matters: This interpretable ML framework transforms sleep disorder risk stratification for seniors with multimorbidity, enabling precision interventions and improved geriatric care.

A collaboration between Hunan Provincial People’s Hospital and Monash University demonstrates that increased dietary lutein and zeaxanthin intake correlates with reduced biological age acceleration across cardiovascular, hepatic, and renal systems. This conclusion stems from statistical analysis of NHANES 2007–2015 cohort data, complemented by transcriptomic investigations of telomere regulation and inflammatory pathways.

Key points

• High combined lutein/zeaxanthin intake significantly reduces biological age acceleration across cardiovascular, renal, and hepatic systems in NHANES cohort.
• Cox regression shows Q4 LZ intake lowers all-cause mortality risk by ~40–45% compared to Q1.
• Transcriptomic profiling identifies telomere maintenance, metabolic reprogramming, and inflammation suppression as lutein’s anti-aging mechanisms.

Why it matters: This finding highlights a noninvasive, dietary strategy to decelerate organ‐specific aging, offering scalable interventions for healthy longevity.

Researchers from Near East University and collaborating institutions analyse how theoretical and practical AI knowledge influences primary school teachers’ sustainable integration of AI in Northern Cyprus. Using a structured survey and Structural Equation Modeling, they demonstrate that teachers’ beliefs and attitudes critically mediate the impact of AI knowledge on classroom adoption. These findings underscore the importance of professional development that fosters both AI competence and positive perceptions to secure lasting educational innovation.

Key points

• Teachers’ beliefs and attitudes mediate the impact of both theoretical and practical AI knowledge on sustainable AI integration.
• Structural Equation Modeling on 340 primary teachers’ survey data demonstrates that positive perceptions explain 58% of variance in integration ability.
• Both theoretical and practical AI knowledge contribute indirectly to classroom AI adoption, highlighting the need for integrated conceptual and hands-on training.

Why it matters: By revealing that teachers’ perceptions mediate AI adoption, this research reshapes training strategies for enduring, scalable educational technology integration.

Researchers from the University of Neuchâtel administer 10 % sucrose diets supplemented with 8 mM H₂O₂ or 1 mg/mL ascorbic acid in a 3×3 early–late feeding design. They track individual mosquito lifespan, fecundity after blood meals, and Vavraia culicis spore counts to determine how supplement timing modulates life‐history trade‐offs and parasite tolerance.

Key points

• Early prooxidant intake (8 mM H₂O₂) extends Anopheles gambiae lifespan by ~4–5 days, especially in uninfected mosquitoes.
• Antioxidant supplementation (1 mg/mL ascorbic acid) increases egg production by ~30% irrespective of Vavraia culicis infection.
• Early prooxidant or antioxidant diets raise V. culicis spore loads by ~50%, demonstrating timing‐dependent shifts from resistance to tolerance.

Why it matters: Timing of oxidative interventions reveals new leverage points to disrupt mosquito vector fitness and malaria transmission potential.

A team from the Centre for Global Change at Sol Planje University presents LSTM-GAM-xAI, a hybrid deep learning and generalized additive model enhanced with LIME explainability and causal analysis. It forecasts concentrations of PM2.5, PM10, O₃, NO₂, NO, NOₓ, SO₂, and CO across 5- and 10-day timesteps with lower MSE than benchmarks, for improved regional air quality management.

Key points

• Integrates LSTM deep learning with a generalized additive model layer to capture nonlinear and temporal pollutant dynamics.
• Employs LIME post-hoc explainability to quantify feature contributions (e.g., NO₂, PM₂.₅) for each air pollutant forecast.
• Validates on synthetic Kimberley datasets across 5- and 10-day timesteps, outperforming LSTM, BiLSTM, GRU, BiGRU, 1DCNN, Random Forest, and XGBoost by lowest MSE.

Why it matters: This hybrid explainable AI framework sets a new standard for accurate, interpretable air quality forecasts, empowering data-driven environmental policy and health protection.

A cross-sectional NHANES study led by Nantong University and Shanghai Lida University shows anti-inflammatory diets significantly reduce phenotypic age acceleration and mitigate adverse effects of vigorous-intensity exercise on biological aging.

Key points

• Anti-inflammatory diets reduce phenotypic age acceleration by up to 2.72 years compared to pro-inflammatory diets.
• Sufficient vigorous physical activity alone increases PhenoAgeAccel by 0.81 years, but its pro-aging effects are offset when combined with an anti-inflammatory diet.
• CatBoost machine-learning analysis identifies BMI, DII, gender, age, race, and physical activity as top predictors of biological aging.

Why it matters: Demonstrates how targeted dietary inflammation control can modulate biological aging and improve longevity outcomes.

Researchers from KIIT University, University College Dublin, ICAR and Anglia Ruskin University review how AI-driven methods such as machine learning, federated learning and computer vision tailor nutritional strategies to individual biological profiles. The study also examines AI applications in food manufacturing—predictive maintenance, quality control and waste minimization—to enhance resilience and sustainability in food systems. Key ethical, privacy and explainability challenges are discussed alongside pathways for clinical and industrial integration.

Key points

• Supervised and reinforcement learning models predict individual glycemic responses, reducing postprandial excursions by up to 40%.
• CNN-based image recognition (e.g., YOLOv8, vision transformers) achieves >90% accuracy in food classification for real-time nutrient estimation.
• Federated learning frameworks with secure aggregation enable privacy-preserving multi-center health data analytics under GDPR/HIPAA compliance.

Why it matters: By uniting AI-driven personalization and sustainable manufacturing, this review charts transformative pathways for precision nutrition and resilient food systems.

Researchers at Michigan State University assess how the exposome—cumulative environmental and dietary exposures—modulates oxylipin metabolism via CYP, LOX, COX, and epoxide hydrolase pathways. By detailing molecular links between vitamins, metals, PUFAs, and lipid mediators, they highlight mechanisms influencing cellular senescence and inflammation to improve healthspan.

Key points

• Exposome factors (vitamins A, D, E, K; trace metals; PUFAs) modulate oxylipin profiles influencing senescence and inflammation.
• CYP450, COX, and LOX enzymes produce epoxy- and hydroxy-PUFAs; sEH regulates their bioactivity, impacting healthspan.
• Targeting exposome–lipid interactions (e.g., sEH inhibition) offers therapeutic avenues to extend healthy aging in preclinical models.

Why it matters: Mapping exposome–oxylipin links uncovers modifiable metabolic pathways, guiding novel strategies to extend healthy human lifespan.

Researchers from Universiti Putra Malaysia employ CiteSpace and VOSviewer to analyze 450 Web of Science articles on AI-assisted psychological interventions for stroke survivors, mapping collaboration networks, publication trends, and emerging hotspots such as ischemic stroke and anxiety management.

Key points

• Dataset of 450 WoSCC articles (2000–2024) analyzed via CiteSpace and VOSviewer
• Calabro Rocco Salvatore leads authorship (9 publications) and McGill University leads institutions (10 publications)
• Emerging research hotspots include ischemic stroke, anxiety, and cognitive impairment in AI-supported care

Why it matters: This bibliometric study highlights evolving AI applications in stroke psychology research, guiding targeted intervention development and interdisciplinary collaborations.

A team from the University of Edinburgh’s Centre for Electronic Frontiers employs AI-driven workflows across five key pillars—materials discovery, device design, circuit synthesis, testing, and digital twin modeling—to accelerate nanoelectronics development, boost yield, and promote greener manufacturing processes.

Key points

• AI-driven materials discovery predicts novel, sustainable nanoelectronic compounds using machine learning surrogate models.
• Advanced neural networks optimize nano-device architectures and automate circuit synthesis, improving performance and reducing design iterations.
• Physics-informed digital twins enable real-time device modeling and predictive maintenance across the electronics supply chain.

Why it matters: This integrated AI framework reshapes nanoelectronics by cutting development cycles, driving sustainable manufacturing, and enabling next-generation device performance.

Researchers at West University of Timișoara investigate AI-induced technostress using the Technostress Creators scale and DASS-21 questionnaires among 217 Romanian adults. Employing structural equation modeling, they demonstrate significant positive associations between AI-related stressors—overload, invasion, complexity, and insecurity—and symptoms of anxiety (β=0.342) and depression (β=0.308), accounting for 11.7% and 9.5% of variance, respectively.

Key points

• Latent technostress construct comprises five factors with loadings: overload (.809), invasion (.813), complexity (.503), insecurity (.735), uncertainty (.314).
• SEM shows technostress predicts anxiety (β=.342, p<.001, R2=.117) and depression (β=.308, p<.001, R2=.095) in a 217-participant Romanian sample.
• Technostress and DASS-21R scales demonstrate strong internal consistency (Cronbach’s α>0.80) across all measured dimensions.

Why it matters: By quantifying how AI-induced technostress contributes to anxiety and depression, this study highlights urgent mental health implications as AI integrates into everyday life.

A team at Beijing Jiaotong University examines how organizational AI integration enhances employee knowledge sharing by creating learning opportunities. Surveying 364 employees, structural equation modeling reveals that paradoxical leadership and technophilia positively moderate the indirect effect of AI adoption on knowledge exchange, offering evidence-based guidelines for managers.

Key points

• AI adoption directly increases learning opportunities (β=0.169, p<0.001) in SEM analysis of 364 employees.
• Learning opportunities mediate the AI–knowledge sharing link with an indirect effect of 0.047 (95% CI[0.030,0.066]).
• Paradoxical leadership and technophilia significantly strengthen both the AI–learning relationship (β=0.119, p<0.001; β=0.045, p<0.05) and the downstream knowledge-sharing pathway.

Why it matters: By identifying learning opportunities, leadership style, and technophilia as key drivers, this research offers strategies to maximize AI-driven collaboration.

Researchers at the European University of Rome and partner institutions analyze AI adoption by assessing anxiety, usage, positive attitudes, and perceived knowledge among 335 adults. They find women report higher AI anxiety and lower AI engagement, with gender moderating anxiety’s effect on attitudes.

Key points

• Women report significantly higher AI anxiety and lower positive attitudes toward AI, perceived knowledge, and use (MANOVA η²=0.162).
• PROCESS moderation analysis shows gender moderates the negative relationship between AI anxiety and positive AI attitudes, with anxiety impacting men more steeply.
• Prior AI use positively predicts attitudes (β>0), while age and perceived AI knowledge have no direct effect.

Why it matters: Identifying gender-specific AI apprehensions and engagement patterns informs interventions to bridge the AI adoption gap and promote inclusive digital policy.

A team from Nanjing Audit University investigates how Big Five personality traits influence static and dynamic trust in AI-driven drone missions across PC and VR modalities. Using a D3QN-based UAV simulation in Unity, they measure trust before and after interaction to inform adaptive, personality-aware human–machine interface designs.

Key points

• Unity-based UAV surveillance simulation uses D3QN for autonomous path planning and obstacle avoidance.
• Chinese TIPI questionnaire measures Big Five traits; extroversion and emotional stability highlighted.
• Static trust (T0) assessed pre-interaction; dynamic trust (T1) measured post-interaction on PC and VR.
• Extroversion significantly predicts initial trust; emotional stability enhances post-interaction trust in PC.
• Static trust consistently predicts dynamic trust across modalities, explaining up to 21.9% of T1 variance.
• VR yields higher initial trust, while PC delivers greater dynamic trust, per independent t-tests.

Why it matters: By revealing static trust as the foundation for evolving human-machine trust and identifying extroversion and emotional stability as key drivers, this study guides the design of adaptive, user-centric AI systems. Tailoring interfaces to individual personalities can enhance safety, reliability, and long-term engagement in AI applications.

Frontiers in Psychology presents a breakthrough in physical education, showcasing an AI-driven system that uses markerless motion capture and real-time data analysis. Similar to personalized digital coaching, this framework refines student performance through closed-loop feedback mechanisms, offering a promising method for enhancing both engagement and health outcomes in educational settings.

The article provides insights into the evolving dynamics of human-AI interaction, demonstrating how various agents—robots, avatars, and chatbots—transform social exchanges. Using real-life analogies, Albert Łukasik’s 2025 study reveals that design nuances affect user trust and emotional responses, such as when AI companions foster comfort during isolation.

Frontiers in Psychology presents a detailed 2025 study by Xin Xin on balancing functional efficiency with aesthetic design in service robots. The research argues that incorporating human-like features can enhance social interaction and practical usability, offering a clear framework for how design choices influence user engagement. This study provides compelling insights for those interested in the future of robot design.

A 2025 study by Guanglu Xu in Frontiers in Psychology reveals that proactive personality in migrant workers is linked to lower technical unemployment risk. Despite AI learning alone not reducing risk, enhanced self-efficacy mediates this relationship, suggesting that proactive behavior and confidence are key during AI-driven changes.

A recent study by DongLi Ma in Frontiers in Psychology introduces HCM-Net, a hierarchical deep learning framework combining EEG signal analysis, graph neural networks, and LSTM to quantify crime motivation. The work also introduces DRAS for dynamic risk adaptation, providing a promising use case in forensic psychology.