To understand the significance of the proposal, it helps to recall what mitoxyperilysis actually is. The term describes a recently defined mode of regulated cell death in which the fatal event is the rupture of the plasma membrane, the outer envelope that keeps a cell intact. What distinguishes mitoxyperilysis from other lytic deaths such as pyroptosis or necroptosis is the upstream cause: the membrane fails because mitochondria, the cell’s energy-producing organelles, generate reactive oxygen species that chemically attack nearby membrane lipids. This process, known as lipid peroxidation, degrades the phospholipids that give the membrane its structural integrity. Once enough lipid damage accumulates, the membrane destabilizes and bursts, releasing the cell’s contents. Crucially, the damage is not uniform; it is spatially restricted, concentrated at sites where mitochondria linger close to the plasma membrane. The new review takes that spatial restriction seriously and asks what molecular machinery determines where mitochondria sit in the first place.

The answer proposed by Zhenhong He, Yishu Zhong, Yuhan Luo, Hanyu Li, Yingying Yu and Dingsu Bao centers on mTORC2, one of two large protein complexes built around the mechanistic target of rapamycin kinase. While its sibling complex mTORC1 is famous as a nutrient and growth sensor, mTORC2 is best known for controlling the cytoskeleton, the dynamic scaffold of actin filaments and associated motors that gives cells their shape and drives their movement. mTORC2 promotes the activity of RhoA, a small GTP-binding protein that orchestrates actin remodeling, and supports the formation of lamellipodia, the sheet-like actin protrusions that cells extend when they migrate or reshape their surfaces. According to the framework, when immunometabolic stress keeps mTORC2 switched on for prolonged periods, this cytoskeletal program is disrupted rather than enhanced. Sustained mTORC2 activation suppresses RhoA-dependent cytoskeletal dynamics and lamellipodium formation, leaving the cell’s outer architecture stiff and poorly remodeled.

This suppression has a spatial consequence that lies at the heart of the model. In a healthy, actively remodeling cell, mitochondria are constantly repositioned along cytoskeletal tracks, distributed to wherever energy demand is highest and kept away from regions where their reactive byproducts could do harm. If RhoA-driven dynamics and lamellipodial activity falter, the review argues, mitochondria lose this mobility and form prolonged contacts with the plasma membrane. The organelles effectively become parked against the cell’s outer wall. Because mitochondria continuously generate superoxide and other reactive oxygen species as byproducts of oxidative phosphorylation, any ROS they release at these contact sites accumulates locally rather than being diluted across the cytoplasm. The result is a microenvironment of intense oxidative stress confined to the narrow gap between the mitochondrial outer membrane and the inner face of the plasma membrane, precisely where lipid peroxidation can do the most structural damage.

The chemistry that follows is well characterized in adjacent fields. Reactive oxygen species abstract hydrogen atoms from the unsaturated fatty acid chains of membrane phospholipids, initiating chain reactions of peroxidation that render lipids rigid, fragmented and prone to forming pores. Ferroptosis researchers have shown how extensively lipid peroxidation can compromise membrane integrity when it overwhelms cellular repair systems such as glutathione-dependent peroxidases. The mitoxyperilysis framework borrows this chemistry but adds a crucial geometric constraint: because the peroxidizing ROS are generated at mitochondria held in place near the membrane, the lipid damage is delivered to a specific patch of the plasma membrane rather than arising randomly. Localized peroxidation destabilizes that patch, and once a critical threshold of damage is crossed, the membrane lyses. In this view, cell death is not merely a matter of how much oxidative stress a cell experiences, but of where that stress is delivered, a distinction the authors argue has been underappreciated in the redox cell death literature.

A substantial portion of the review is devoted to positioning mitoxyperilysis against the growing family of redox-associated death modalities. Ferroptosis, the best-studied example, depends on iron-catalyzed lipid peroxidation and can be blocked by lipophilic antioxidants and iron chelators, but its lipid damage is generally considered a cell-wide phenomenon shaped by lipid composition and repair capacity. Pyroptosis and necroptosis rupture the membrane through pore-forming proteins such as gasdermins and MLKL respectively, with oxidative stress playing at most a modulatory role. Mitoxyperilysis, by contrast, is defined by its spatial signature: oxidative membrane damage that is restricted to mitochondrial contact zones. The authors emphasize that this defining feature, spatially restricted oxidative membrane damage, is what sets the modality apart and justifies treating it as a distinct conceptual category rather than a variant of ferroptosis. The comparison also highlights a broader principle emerging in cell death research, namely that subcellular organization, not just molecular identity, shapes how cells die.

The framework also carries implications for disease. Chronic immunometabolic stress, the simultaneous burden of inflammatory signaling and metabolic dysregulation, is a hallmark of conditions ranging from obesity-associated inflammation and atherosclerosis to chronic joint and tissue degeneration. The authors, who work in an orthopedics department and a college of integrated traditional Chinese and Western medicine, suggest that tissues subjected to sustained immunometabolic stress may be prone to mitoxyperilysis-like injury, in which persistently active mTORC2 immobilizes mitochondria near membranes and primes localized oxidative attack. If the model holds, it could help explain patterns of tissue damage that current death classifications do not fully account for, and it points toward therapeutic strategies that are spatially targeted rather than globally suppressive. Inhibiting mTORC2, restoring RhoA-dependent cytoskeletal dynamics, or enhancing mitochondrial mobility might each prevent the fatal apposition of mitochondria and membrane without shutting down mitochondrial metabolism altogether.

It is important to read the review for what it is: a conceptual framework, explicitly framed with cautious language. The authors repeatedly use terms such as may act, could promote and may facilitate, signaling that the causal chain from sustained mTORC2 activation through cytoskeletal suppression to mitochondrial immobilization and localized peroxidation has not yet been demonstrated end to end in living cells. Key experimental tests remain open, including direct imaging of mitochondria-plasma membrane contact duration under mTORC2 hyperactivation, quantitative mapping of lipid peroxidation products at contact sites, and genetic or pharmacological rescue experiments showing that restoring cytoskeletal dynamics prevents lytic death. The work was supported by the Sichuan Provincial Administration of Traditional Chinese Medicine and by a science and technology cooperation program between the Luzhou Municipal People’s Government and Southwest Medical University, and the authors declare no competing interests.

Even as a hypothesis, the framework is likely to resonate because it reframes a familiar molecule in an unfamiliar role. mTORC2 has long been studied as a signaling hub for growth, survival and cytoskeletal organization, and mTOR inhibitors are staples of cancer and immunology research. The idea that the same complex could function as a spatial regulator of cell death, effectively deciding where a cell’s own mitochondria are allowed to sit, adds a geometric dimension to signaling biology that few models have attempted. It also joins a broader movement in cell biology that treats the position of organelles as information in its own right, as important as their enzymatic activity. If subsequent experiments confirm the mTORC2-cytoskeleton-mitochondrial positioning axis, mitoxyperilysis could become a textbook example of how subcellular architecture converts metabolic stress into a precisely located lethal blow, and the search for spatially targeted therapies in immunometabolic disease would gain a concrete molecular starting point.

Subject of Research: An mTORC2-driven mitochondrial positioning framework links immunometabolic stress to spatially restricted oxidative membrane damage in mitoxyperilysis

Article Title: An mTORC2-driven mitochondrial positioning framework links immunometabolic stress to spatially restricted oxidative membrane damage in mitoxyperilysis

Article References: He, Z., Zhong, Y., Luo, Y., Li, H., Yu, Y., & Bao, D. (2026). An mTORC2-driven mitochondrial positioning framework links immunometabolic stress to spatially restricted oxidative membrane damage in mitoxyperilysis. Cell Death Discovery. https://doi.org/10.1038/s41420-026-03362-x

Image Credits: AI Generated

DOI: 10.1038/s41420-026-03362-x

Keywords: mTORC2-driven, mitochondrial, positioning, framework, links, immunometabolic, stress, spatially, restricted, oxidative, membrane, damage

The authors begin from an uncomfortable observation: decades of behavior change research, from the theory of planned behavior to self-determination theory and socioecological models, explain only a modest proportion of the differences in physical behavior between and within individuals. People are notoriously inconsistent—one week they jog daily, the next they barely leave the sofa. The researchers attribute this weak predictive power to three structural shortcomings in existing models. First, most frameworks focus either on the short-term determinants that drive behavior or on the long-term health consequences that flow from it, rarely both. Second, they overlook the fact that consequences can loop back and become determinants: fitness gained through months of exercise also shapes how active you feel like being tomorrow. Third, and most fundamentally, they ignore the arithmetic of the day itself.

That arithmetic is unforgiving. Sleep, sedentary behavior, and physical activity are locked together within the fixed 1,440 minutes available, so every additional minute spent in one behavior must come at the expense of another. A person who adds thirty minutes of morning running necessarily removes thirty minutes from sleep, sitting, or some other activity. Traditional single-behavior studies, which examine physical activity in isolation, cannot capture these trade-offs. The new model builds on earlier 24-hour frameworks—including Pedišić’s Activity Balance Model, the Framework for Viable Integrative Research in Time-Use Epidemiology, and Rosenberger and colleagues’ 24-hour Activity Cycle—but goes further by explicitly embedding the full daily composition within a dual-process psychological architecture and a health-outcome architecture, connected by eight distinct pathways labeled A through H.

The left side of the model, adapted from the neurocognitive affect-related model, describes how the brain’s moment-to-moment machinery shapes what we do with our time. Its central constructs are affective states and executive functions. Following the work of Duncan and Barrett, the authors treat affect as a neurophysiological state defined by two dimensions: valence, the pleasantness or unpleasantness of a feeling, and energetic arousal, the level of activation. Executive functions, as defined by Diamond, encompass inhibition control, working memory, and cognitive flexibility—the mental capacities that let us plan, resist temptation, and stay focused on long-term goals despite distractions. The model proposes that optimal daily compositions of sleep, sitting, and activity can enhance these executive functions, which in turn foster more positive affective responses, which then facilitate healthier future behavior compositions.

Crucially, these pathways are reciprocal. Pathway A suggests that executive-function-based cognition shapes the emotional responses induced by exercise or prolonged sitting; pathway B connects those emotional responses to the future balance of daily behaviors, since affective reactions during a workout predict whether a person returns to exercise. Pathways C and D close the loop: cognitive preparation such as time management and goal setting influences the day’s activity composition, while the composition itself—regular exercise, adequate sleep—feeds back to improve cognitive function. The authors also acknowledge bidirectionality in the reverse direction, noting that physical behavior can predict momentary affect and that momentary affect can predict executive function performance, making the entire left side of the model a dynamic, self-modifying system rather than a one-way causal chain.

The right side of the model, adapted from the health model of Bouchard, Blair, and Haskell, traces the physiological consequences. Here the claim is that an optimal daily composition improves health-related fitness markers such as cardiorespiratory fitness (pathway E), which then improves the state of physical and mental health, from well-being to mortality risk (pathway F). The evidence base for these links is substantial. A systematic review of fifty-six isotemporal substitution studies concluded that reallocating time between sleep, sedentary behavior, light-to-moderate activity, and moderate-to-vigorous activity is associated with numerous health outcomes. A meta-analysis further showed that short sleep is significantly linked to mortality, diabetes, cardiovascular disease, coronary heart disease, and obesity. These pathways, too, can run backward: fitness markers predict individual behaviors, health status predicts fitness, and illness or injury—captured by pathway H—can abruptly reshape the whole day, as anyone with a broken leg who suddenly spends far more time sleeping and sitting can attest.

What distinguishes this framework from its predecessors is its insistence on temporal resolution. The authors divide the eight pathways into short-term dynamics operating within days, medium-term associations across weeks and months, long-term effects over years, and mixed pathways spanning days to years. Affective states fluctuate hour to hour and even minute to minute, so pathways A, B, and C demand high-granularity, real-world data. The state-of-the-art tool here is ambulatory assessment: smartphone-based electronic diaries capturing real-time self-reports, paired with wearable sensors measuring the full 24-hour behavior composition objectively and repeatedly. This approach preserves ecological validity and avoids the retrospective biases that plague questionnaire studies. By contrast, the fitness pathway E requires weeks or months of observation, ideally through randomized interventions with repeated and follow-up measurements, while the health pathways F and G unfold over years and call for long-term cohort studies—one 18-year community study found that habitual physical activity related positively to both fitness and health status decades later.

The model also arrives at a moment when the statistical machinery for 24-hour research has matured. Compositional data analysis treats sleep, sedentary time, and activity as parts of a finite whole, analyzing them relative to one another rather than as independent variables, and the compositional isotemporal substitution method estimates what happens to a health outcome when a fixed block of time is shifted from one behavior to another. The empirical payoff is already visible. In an analysis of six cross-sectional studies covering 15,253 participants, moderate-to-vigorous physical activity showed the strongest and most time-efficient protective associations with cardiometabolic outcomes, while sedentary behavior was the only behavior with clearly adverse associations regardless of duration. The Maastricht Study, with 2,388 participants, found that less sitting and more standing, physical activity, and sleep were associated with better cardiometabolic health and glycaemic control. Device-based measurement via wearables now makes it possible to capture every facet of the daily composition with time-stamped precision, and the model’s components can be refined as needed—splitting activity into standing, light, and vigorous categories, or sleep into REM and non-REM stages, or even distinguishing outdoor from indoor activity.

The long-term ambition is strikingly concrete: to identify the optimal balance of 24-hour behaviors that maximizes health benefits and promotes longevity and well-being. Tremblay and colleagues anticipate that the 24-hour approach will eventually underpin individualized, precision movement guidelines tailored to personal characteristics and circumstances—several countries and the World Health Organization already issue integrated 24-hour movement guidelines. The authors are careful to note that the relationships in their model are shaped by genetic predispositions, social and physical environments, and individual circumstances, which may moderate, mediate, or confound the pathways, so generalization demands caution. But by combining behavioral, affective, and health determinants and consequences in a single flexible framework, and by specifying exactly which timescale each hypothesis should be tested on, the 24-hour cognitive-affective physical behavior model gives researchers something previous theories did not: a shared map of the entire day, drawn to scale, with the clock built in.

Subject of Research: The 24-hour cognitive-affective physical behavior model: a theoretical framework for studying determinants and health consequences of physical activity, sedentary behavior, and sleep

Article Title: The 24-hour cognitive-affective physical behavior model: a theoretical framework for studying determinants and health consequences of physical activity, sedentary behavior, and sleep

Article References: Giurgiu, M., & Ebner-Priemer, U. W. (2025). The 24-hour cognitive-affective physical behavior model: a theoretical framework for studying determinants and health consequences of physical activity, sedentary behavior, and sleep. Journal of Activity, Sedentary and Sleep Behaviors, 4(1), Article 6. https://doi.org/10.1186/s44167-025-00077-9

Image Credits: AI Generated

DOI: 10.1186/s44167-025-00077-9

Keywords: hour, cognitive-affective, physical, behavior, model, theoretical, framework, studying, determinants, health, consequences, activity

The framework addresses a problem that has plagued translational medicine for decades: most drugs that look promising in animal models fail in human trials, and many surprises in human research could have been anticipated if animal evidence had been examined more carefully and systematically. Traditional systematic reviews usually focus on a single evidence stream, applying methods honed for clinical trials to human studies and, less often, adapted checklists to laboratory animal experiments. Rarely are the two streams synthesized together with explicit attention to how they agree, conflict or leave gaps. The GALENOS team argues that this separation is particularly costly in early-stage research, where the goal is not to establish definitive treatment effects but to identify and prioritise promising new directions, therapeutic targets and mechanisms of action. Combining human and animal evidence, they contend, can yield insights neither source alone can provide.

At the heart of the framework is a commitment to living evidence. Rather than producing a static review that is outdated before it is published, GALENOS reviews are designed to be continually updated as new studies appear. This is made feasible through automation: the team employs tools such as the Automated Systematic Search Deduplicator to manage the flood of candidate records, and the Systematic Review Facility, or SyRF, to coordinate screening and data extraction, including crowdsourced contributions. Large language models are also being incorporated to support semi-automated screening and data handling, although the authors are careful to position automation as an aid to, not a replacement for, human judgment. The living approach is coupled with co-production: people with lived experience of anxiety, depression and psychosis help shape research questions, interpret findings and judge which uncertainties matter most to patients, ensuring that prioritisation reflects relevance as well as statistical strength.

Methodologically, the framework builds on established best practices in systematic review methodology while introducing novel components. Research questions are framed using PICO elements, defining population, intervention, comparator and outcome, but with careful translation across species and study designs, since an intervention in a mouse model of psychosis and an antipsychotic trial in humans must be mapped onto a common conceptual structure before their results can be compared. For animal studies, the team applies risk-of-bias tools derived from the SYstematic Review Centre for Laboratory animal Experimentation and reporting standards aligned with the ARRIVE guidelines, addressing the well-documented problems of small sample sizes, publication bias and methodological laxity in preclinical research. Human studies are appraised with conventional Cochrane-style tools, and both streams are assessed for the risk of bias arising from missing evidence using the ROB-ME approach.

Quantitative synthesis within the framework handles the awkward fact that human and animal studies measure outcomes on different scales and with different statistics. The authors describe how effect measures such as mean differences, standardised mean differences, risk ratios and odds ratios can be computed within each evidence stream, and how the normalised mean difference can bridge human and animal outcomes onto a comparable metric. Random-effects meta-analysis, using restricted maximum likelihood estimation, pools studies within each species and design category. Crucially, synthesis is not limited to pairwise comparisons; the framework accommodates networks of interventions and mechanisms, allowing reviewers to ask not only whether a target works but how it performs relative to alternatives across both human and animal data.

The most distinctive innovation is what the team calls the Summary of Evidence Table, or SET. A SET is a structured, transparent container that assembles the synthesized findings from human and animal studies on a given question, alongside appraisals of their credibility, consistency and biological plausibility. On top of the SETs sits the GALENOS Approach to Triangulating Evidence, or GATE, a formal procedure for integrating the streams. Rather than simply averaging across species, GATE asks structured questions: do human and animal findings agree in direction and magnitude, does one stream fill gaps the other cannot, and where they conflict, what biases or translational limitations best explain the discrepancy? The output is a graded judgement about the maturity of the evidence, which in turn feeds directly into research prioritisation decisions about which targets warrant clinical testing, which need better animal work, and which should be deprioritised.

The framework is not merely theoretical. The GALENOS consortium maintains living systematic reviews on concrete therapeutic questions, and the paper illustrates its application with two examples. One concerns trace amine-associated receptor 1, or TAAR1, agonists for psychosis, a target class that has recently progressed to clinical testing; the living review collates animal evidence on receptor mechanisms alongside emerging human trial data to judge whether the translational case remains sound. The other concerns pro-dopaminergic interventions for anhedonia, the blunting of pleasure and motivation that pervades depression and schizophrenia, where animal work on dopaminergic circuitry must be reconciled with early human studies. In both cases, the SET and GATE machinery makes explicit how strong each strand of evidence is, where uncertainty concentrates and what future experiments would be most informative.

Independent experts in evidence synthesis are likely to welcome the framework’s insistence on transparency. By laying out, in tabular form, exactly what is known from each species, how credible each finding is, and where the streams converge or diverge, the approach makes prioritisation decisions auditable rather than opaque. This matters in mental health particularly, where the translational failure rate has been devastatingly high and where the DSM-based grouping of patients may obscure biological heterogeneity that animal models capture differently. The authors are candid about the epistemological challenges: animal models are approximations, human diagnostic categories are constructs, and integrating them involves assumptions that must be stated, not hidden. The framework’s contribution is precisely to force those assumptions into the open.

The implications extend beyond psychiatry. The general architecture of co-produced questions, living systematic reviews, source-specific synthesis, and structured triangulation through SETs and GATE could be applied in any field where preclinical and clinical evidence must be weighed together, from neurology to oncology. For funders and research commissioners, the framework offers a defensible mechanism for deciding where scarce research money goes; for scientists at the bench, it offers a clearer picture of which of their findings the clinical world considers robust; for patients and people with lived experience, it offers a seat at the table where research agendas are set. The GALENOS team presents the approach as a working system under continuous refinement, with its living reviews and prioritisation outputs publicly accessible, inviting the research community to test, adapt and challenge it as the evidence accumulates.

Subject of Research: A framework for collating and synthesizing evidence from human and animal studies

Article Title: A framework for collating and synthesizing evidence from human and animal studies: the GALENOS approach

Article References: Chiocchia, V., Higgins, J. P., Siafis, S., Macleod, M., Thomas, J., Furukawa, T. A., Downs, J., Cipriani, A., Egger, M., & Salanti, G. (2026). A framework for collating and synthesizing evidence from human and animal studies: the GALENOS approach. BMC Medicine. https://doi.org/10.1186/s12916-026-05205-z

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05205-z

Keywords: GALENOS, evidence synthesis, living systematic reviews, triangulation, animal studies, human studies, mental health, research prioritisation, TAAR1, co-production, framework, collating

The study, conducted by Fei Song and Michael Dunn of the Centre for Biomedical Ethics at the National University of Singapore together with Zhenni Hu of Nanjing Normal University, examined 43 articles indexed in the Chinese Social Sciences Citation Index along with four policy reports. The analysis reveals that Chinese scholars have made substantial contributions to policy and governance discussions around trustworthy AI, but that these contributions remain conceptually scattered, with different disciplines approaching the question of trust from radically different angles. The researchers set out to synthesize this dispersed knowledge into a coherent conceptual account, and in doing so they arrived at a framework that treats trustworthiness not as a property of algorithms alone but as an emergent condition spanning technology, institutions, and human relationships.

The first layer of the framework concerns technical foundations. This is the layer most familiar to engineers and computer scientists, encompassing properties such as reliability, robustness, transparency, explainability, fairness, and safety. Chinese-language scholarship in this domain is extensive and spans a striking range of applications. Researchers have examined explainable learner models as the technical key to trustworthy personalized learning, explored blockchain-driven approaches to financial security intelligence, and investigated how algorithmic transparency in personalized recommendation systems shapes users’ perceptions of trustworthiness. Work on data annotation governance has highlighted backstage risks in the labor-intensive processes that feed machine learning systems, while archival science perspectives have been brought to bear on the question of algorithm provenance, asking how the origins and histories of algorithms can be documented and verified. These contributions demonstrate that the technical layer of trustworthiness is not merely a matter of model architecture but extends deep into the data pipelines, documentation practices, and supply chains that underpin AI systems.

The second layer addresses governance and regulatory arrangements. Here the Chinese literature is particularly distinctive, reflecting the country’s active policy discourse on AI regulation. Scholars have assessed pathways for trustworthy AI legislation, explored innovation-friendly regulatory designs that balance safety with technological development, and analyzed governance practices in terms of models, actors, objects, and tools. Policy documents play an important role in this layer, including the Ministry of Science and Technology’s 2019 governance principles for developing responsible AI, a white paper on trustworthy artificial intelligence produced by the China Academy of Information and Communications Technology, and more recent industry ecosystem reports and governance surveys. The review shows that Chinese scholars have been grappling with the same fundamental questions that animate regulators in Europe and the United States: how to translate abstract ethical principles into enforceable rules, how to assign liability when AI systems fail, and how to design oversight mechanisms that keep pace with rapidly evolving technology. One striking example is legal scholarship on liability when human–machine co-piloting fails, examining the negligence of safety operators through the lens of the principle of reliance.

The third layer, and the one the authors argue is most often neglected in purely technical accounts, concerns the conditions of human–AI interaction under which trust is formed and calibrated. The Chinese-language literature here is remarkably rich, drawing on empirical psychology, communication studies, and philosophy. Researchers have studied trust in automated vehicles, safety trust in intelligent domestic robots, and human–AI mutual trust in the era of artificial general intelligence. A dual-pathway model of trust calibration distinguishes between trust dampening and trust promoting mechanisms, offering a nuanced picture of how users adjust their reliance on machines. Other studies examine how anthropomorphism affects trust in AI agents, with a meta-analysis synthesizing evidence on how humanlike features in AI systems influence trust and how contextual factors moderate that effect. Work on large language models has explored how machine hallucinations impair human–machine trust and what recalibration mechanisms might restore it, a question of obvious urgency as chatbots and AI assistants become ubiquitous.

What emerges from this synthesis is the insight that these three layers are not independent. An AI system can be technically excellent, with high accuracy and robust performance, and still fail to be trustworthy if the governance arrangements around it are inadequate or if the conditions of human interaction systematically miscalibrate trust. Conversely, strong institutions and well-designed interaction conditions cannot compensate for fundamentally unreliable technology. The authors’ central claim is that an AI system is trustworthy only if it meets the requirements at all three layers simultaneously. This conjunctive structure has practical implications: it means that trustworthiness cannot be certified by any single test, audit, or label, but requires integrated assessment across the full stack from model internals to regulatory environments to the design of the encounters between humans and machines.

The framework also speaks to a long-running philosophical debate about whether trust in AI is even coherent. Some Western philosophers argue that trust, properly understood, requires mutual expectations and moral agency that machines cannot possess, and that what we call trust in AI is really a form of reliance or confidence. Others contend that it is entirely possible to trust medical AI systems and that denying this misunderstands the functional role of trust in clinical decision-making. The Chinese literature adds distinctive voices to this debate, with scholars asking directly whether artificial intelligence can serve as a trustee and defending the notion of trustworthy AI from a logical perspective. The review also engages with a network account of trustworthy AI, which holds that trustworthiness is distributed across the networks of actors, artifacts, and institutions surrounding an AI system rather than residing in the system alone. The three-layer framework can be read as a structured elaboration of this network idea, specifying the technical, institutional, and interactional nodes that such a network must contain.

The methodological significance of the study lies partly in its source base. Scholarship published in Chinese remains largely absent from mainstream academic debates, a gap reinforced by publication incentives that push Chinese humanities and social science researchers toward international English-language venues. By drawing on the Chinese Social Sciences Citation Index, the authors accessed a body of work that includes empirical studies using artificial neural networks and Q methodology to map the dimensions of human–AI trust, philosophical analyses grounded in Confucian traditions, and domain-specific investigations in education, journalism, healthcare, and finance. The result is a more globally representative picture of what the research community actually knows about trustworthy AI. The authors also note distinctive features of the Chinese context, including a practice-oriented view of science and technology that shapes how trustworthiness is framed, and a governance culture in which state-led principles, industry white papers, and academic analysis interact closely.

The study concludes by identifying three directions for future research. First, the conceptual fragmentation of the field calls for continued theoretical work to integrate technical, governance, and interactional perspectives into unified accounts of trustworthiness. Second, the rapid emergence of generative AI and large language models creates new empirical challenges, from hallucination-induced trust erosion to the trust dynamics of human–machine dialogue, that demand updated frameworks. Third, cross-linguistic and cross-cultural comparative research is needed to test whether the three-layer structure holds across different scholarly traditions and regulatory cultures, or whether trustworthy AI means genuinely different things in different contexts. As AI systems increasingly mediate news consumption, medical diagnosis, education, and financial services, the stakes of getting trustworthiness right could hardly be higher. This review makes the case that answering the question well requires listening to conversations that have been happening in Chinese all along, and that the resulting framework, grounded in arguments advanced in the Chinese context, offers a more robust conceptual account of how disparate claims about trust fit together than either technical or governance approaches alone can provide.

Subject of Research: Trustworthy AI beyond the technical: a three-layer framework informed by Chinese-language scholarship

Article Title: Trustworthy AI beyond the technical: a three-layer framework informed by Chinese-language scholarship

Article References: Song, F., Hu, Z., & Dunn, M. (2026). Trustworthy AI beyond the technical: a three-layer framework informed by Chinese-language scholarship. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03366-2

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03366-2

Keywords: Trustworthy, beyond, technical, three-layer, framework, informed, Chinese-language, scholarship, scientific research

The review, conducted by Louis Bourgaux and André Tricot of Université de Montpellier Paul Valéry and Fred Paas of Erasmus University Rotterdam and the University of New South Wales, is one of the most comprehensive syntheses of worked-example research ever assembled. The team screened 2,644 unique records across peer-reviewed literature and grey literature, including a targeted search of dissertations and theses, and ultimately included 85 empirical reports: 82 peer-reviewed articles and 3 doctoral dissertations representing 127 distinct studies and experiments. That scale matters because transfer is a notoriously slippery outcome. Studies vary enormously in how far the “new” problem differs from the training example, in how transfer is measured, and in the populations tested. By cataloguing the evidence systematically, the authors were able to identify which design features of worked examples have been experimentally linked to improved transfer, and under what conditions those features succeed or fail.

The theoretical backbone of the review is the interplay between the architecture of human memory and the demands of novel problems. Cognitive load theory holds that working memory can juggle only a small number of interacting elements at once, so instruction for novices must manage that capacity carefully. Worked examples reduce extraneous load by eliminating blind search through the problem space, freeing cognitive resources to build schemas, the organized knowledge structures that allow experts to recognize problem types and select appropriate solution strategies almost automatically. Transfer, however, demands more than a single polished schema for one familiar problem format. It requires flexible schemas that capture the deep, relational structure of a problem class rather than its surface features. The central design question the review addresses is therefore how worked examples can be engineered so that learners extract those abstract structures instead of memorizing surface routines.

A key organizing move in the review is to sort the design strategies by the type of knowledge they target: factual, conceptual, procedural, and metacognitive. Each type supports transfer differently, and the evidence for each differs in strength and boundary conditions. Strategies aimed at conceptual knowledge, for example, include prompting learners to self-explain why each step works, embedding conceptually oriented explanations within the example, and presenting multiple representations of the same underlying principle. Self-explanation prompts are among the most robustly supported interventions in the entire example-based learning literature: asking students to articulate the principle behind a step forces deeper processing than passive reading, and meta-analytic work has repeatedly confirmed its benefits. The review shows that when such prompts are built into worked examples, learners are more likely to map the solution onto novel problems that share structure but not surface appearance.

Procedural knowledge, by contrast, is addressed through techniques that gradually hand responsibility back to the learner. Completion problems and faded examples, in which some steps of a solution are progressively blanked out until the learner solves entire problems independently, occupy a central place here. The logic is a controlled transition: early full guidance builds a schema, then fading demands retrieval and reconstruction, which strengthens the schema and makes it more accessible in unfamiliar situations. Subgoal labeling, the practice of marking the meaningful phases of a solution rather than presenting it as an undifferentiated stream of algebra, has produced some of the most striking transfer effects in the literature, particularly in statistics and programming. Studies by Richard Catrambone and colleagues demonstrated that learners who studied examples organized around labeled subgoals were substantially better at solving novel problems because they could reason about which subgoal applied rather than matching steps one to one.

The review also highlights strategies that exploit comparison and variability. Presenting multiple worked examples that share deep structure but differ in surface story encourages learners to abstract the common principle, a mechanism well documented in analogical learning research since the classic experiments of Mary Gick and Keith Holyoak. Comparing alternative solution methods for the same problem, a line of work developed by Bethany Rittle-Johnson and Jon Star, fosters procedural flexibility, so students can choose among strategies rather than executing a single memorized one. Meanwhile, varying the surface characteristics of practice examples protects learners from overfitting their knowledge to one context. Importantly, the review stresses boundary conditions: variability and comparison can impose heavy working-memory demands on novices, and the expertise reversal effect means that designs helping beginners can actually impede more advanced learners, who no longer need and may be actively distracted by the same scaffolds.

Erroneous examples, worked solutions containing deliberate mistakes that learners must find and fix, emerge as another promising family of designs. Studying an incorrect solution and diagnosing the error can sharpen conceptual understanding, expose common misconceptions, and train the metacognitive skill of monitoring one’s own work for flaws. Several included studies showed transfer gains when erroneous examples were paired with explanatory feedback or when learners compared correct and incorrect versions side by side. The authors note, however, that error-based designs must be handled carefully: poorly integrated errors can confuse low-prior-knowledge students or be mis-encoded, and the balance between the benefits of error analysis and the risk of encoding wrong procedures depends on learner expertise and the quality of accompanying support.

Metacognitive knowledge receives dedicated attention, reflecting a growing recognition that transfer ultimately depends on learners recognizing, on their own, when a known strategy applies to a new situation. Designs here include prompts that ask students to monitor their understanding, reflect on which principles the example illustrates, and plan how they would approach a related but different problem. The review’s synthesis suggests that worked examples alone tend to build competence within the trained format, but that transfer across formats is amplified when the examples explicitly invite learners to think about their own thinking, whether through embedded reflection questions, strategy comparisons, or faded sequences that require self-assessment before support is reintroduced. Collaboration adds another layer: several studies found that pairs discussing worked examples, particularly when knowledge is unevenly distributed between partners, elaborated more deeply and transferred more than individuals working alone.

For practitioners, the payoff of the review is a practical checklist for designing worked examples that promote transfer rather than mere step-following. The checklist distills the evidence into concrete design decisions: clarify which knowledge type the example should target; add subgoal labels to reveal solution structure; insert self-explanation prompts at meaningful points; use faded steps to taper guidance as competence grows; include varied surface features and, where expertise allows, comparison of multiple solutions; consider erroneous examples with adequate support; and calibrate all of this to the learner’s prior knowledge, reducing guidance as expertise develops. The authors emphasize that no single feature is a silver bullet; the strategies interact, and the appropriate combination shifts across the learning trajectory. Still, the checklist gives textbook authors, learning-platform developers, and classroom teachers an evidence-based starting point far more specific than the generic advice to “provide examples.”

The broader significance of the work lies in reframing the debate between guided and minimally guided instruction. Critics of worked examples have argued that heavily guided learning produces brittle knowledge that dies at the classroom door. The 127 studies synthesized here tell a more nuanced story: guidance is not the enemy of transfer, but poorly designed guidance is. When worked examples are engineered to highlight deep structure, demand active processing, and fade strategically, they can produce knowledge flexible enough to survive the trip into unfamiliar territory. In an era when adaptive tutoring systems and AI-assisted learning platforms must decide moment by moment how much help to give, an integrative map of when and how worked examples foster transfer is precisely the tool the field has lacked. The review, published as Volume 38, article 118 of Educational Psychology Review, consolidates that map and hands it to both researchers and practitioners.

Subject of Research: Worked Examples and Transfer: An Integrative Review, Design Framework, and Practitioner Checklist

Article Title: Worked Examples and Transfer: An Integrative Review, Design Framework, and Practitioner Checklist

Article References: Bourgaux, L., Tricot, A., & Paas, F. (2026). Worked Examples and Transfer: An Integrative Review, Design Framework, and Practitioner Checklist. Educational Psychology Review, 38(1), Article 118. https://doi.org/10.1007/s10648-026-10206-8

Image Credits: AI Generated

DOI: 10.1007/s10648-026-10206-8

Keywords: Worked, Examples, Transfer, Integrative, Review, Design, Framework, Practitioner, Checklist, scientific research

The core idea is deceptively simple. Rather than classifying seaweeds only by taxonomy or by their current commercial status, the framework characterizes each species by measurable functional traits: morphological features such as thallus size and branching, physiological features such as photosynthetic rates and nutrient uptake kinetics, and life-history features such as reproductive strategy and growth seasonality. In terrestrial agriculture, this trait-based approach has underpinned decades of crop improvement and ecological screening, allowing scientists to predict how a plant will perform under specific conditions without having to trial every individual genotype. The authors contend that seaweed science is now mature enough to apply the same logic, turning a scattered body of species-specific observations into a coherent, comparative discipline.

The need for such a framework becomes clear when one considers the structure of the seaweed farming industry. Global production is overwhelmingly concentrated in a handful of genera, including the kelp Saccharina and Undaria, the red algae Pyropia, Eucheuma, Kappaphycus, and Gracilaria, and the green alga Ulva. These cultivated species were largely domesticated from wild populations selected for availability and market demand rather than for optimized performance. Meanwhile, thousands of seaweed species remain uncultivated, many of which may possess traits, such as tolerance to warmer water, resistance to disease, rapid growth, or valuable biochemical compositions, that could make them superior candidates for future farming systems, particularly as ocean temperatures rise and cultivation expands into new biogeographic regions.

At the heart of the proposed approach is the recognition that no single trait determines success in aquaculture. A fast-growing species that cannot withstand wave exposure is a poor candidate for open-ocean farms; a hardy species with slow growth may never achieve commercial yields. Trait-based characterization allows these trade-offs to be made explicit. By compiling data on traits such as growth rate, temperature and salinity tolerance, nutrient requirements, epiphyte and pathogen susceptibility, ease of reproduction in hatcheries, and the mechanical properties of the thallus that determine how well a seaweed handles farm infrastructure, researchers can build comparative profiles that reveal which species are most likely to thrive under particular farming conditions and for particular markets.

The framework also speaks directly to one of the most pressing challenges in the field: climate resilience. Many of today’s workhorse species are cultivated near the thermal edges of their tolerance, and marine heatwaves have already caused significant crop losses in several producing regions. Selecting new candidates on the basis of thermal tolerance traits, rather than historical yield alone, could help the industry stay ahead of shifting ocean conditions. Similarly, as interest grows in offshore cultivation, where wave action and depth impose demands that nearshore farms do not, traits such as thallus robustness and attachment strength become decisive selection criteria that conventional species lists simply cannot capture.

Beyond environmental matching, trait-based screening can illuminate a species’ potential end-use. Seaweeds are farmed for wildly different purposes: nori for direct human consumption, carrageenan and agar-producing reds for hydrocolloid extraction, kelps for food ingredients and increasingly for applications in livestock feed and methane reduction. The biochemical profile of a species, its polysaccharide content, protein levels, pigment composition, and secondary metabolites, is itself a set of functional traits. Incorporating these biochemical characteristics into the framework means that candidate selection can be aligned with market opportunities from the outset, rather than discovering after years of agronomic development that a species is agronomically suitable but commercially unattractive.

The authors emphasize that building such a framework requires addressing significant data gaps. Trait information for seaweeds is patchy, fragmented across decades of phycological literature, and often measured with inconsistent methods that make cross-species comparison difficult. For many candidate species, even basic parameters such as maximal growth rate or reproductive biology in cultivation are unknown. The paper calls for coordinated efforts to standardize trait measurements, compile existing data into accessible databases, and prioritize trait characterization for species that show early promise. This is a familiar challenge in the history of trait-based ecology, where the value of comparative frameworks grew directly out of sustained investment in standardized measurement across research groups.

Another important dimension of the framework is its potential to support responsible innovation. Introducing new species into cultivation carries ecological risks, including the possibility of escapes, competition with native flora, and disease transmission. The authors argue that trait-based evaluation can incorporate risk-relevant traits, such as a species’ invasive potential, its reproductive dispersal capacity, and its host status for known seaweed pathogens, alongside performance traits. This integrated screening could help regulators and farmers distinguish between candidates whose domestication is ecologically prudent and those whose traits raise red flags, embedding biosafety into the earliest stages of candidate selection rather than treating it as an afterthought.

The social and economic geography of seaweed farming also figures in the analysis. A large share of global production occurs among smallholder coastal communities in Asia, and any expansion of the industry into new species and new regions will need to respect local capacities, infrastructure, and market structures. Traits such as the simplicity of a species’ cultivation cycle, the availability of seedstock, and the labor intensity of farm operations are directly relevant to whether a candidate species can be adopted by small-scale farmers. A framework that formalizes these practical traits alongside biological ones helps ensure that scientific selection translates into real-world adoption rather than remaining an exercise in laboratory optimization.

Looking forward, the trait-based approach opens doors to more quantitative and even predictive tools. As trait databases grow, statistical and machine-learning methods could identify combinations of traits that predict cultivation success, allowing researchers to screen large numbers of candidate species cheaply before committing resources to farm trials. The same data could guide selective breeding programs within promising species, identifying heritable traits that respond to improvement, much as crop breeders have done on land. In this sense, the framework is not merely a sorting exercise but a foundation for the deliberate domestication of a new generation of seaweed crops, bringing a level of intentionality to aquaculture that has so far been rare in the sector.

The broader significance of the work lies in its reframing of what seaweed farming could become. Today’s industry, for all its impressive growth, rests on a narrow biological base that is vulnerable to disease, climate shocks, and market fluctuations. Diversification is widely recognized as essential, but diversification without a principled selection process risks wasted investment and ecological harm. A trait-based framework offers a shared language for researchers, farmers, investors, and regulators to evaluate candidates systematically, compare trade-offs transparently, and prioritize the species most likely to succeed under the conditions of tomorrow’s oceans. As pressure mounts to expand sustainable marine food production, the ability to choose the right seaweed for the right place and purpose may prove to be one of the most consequential tools the industry acquires.

The analogy with terrestrial crop science is instructive because it highlights how long the payoff horizon for trait-based approaches can be. In agriculture on land, systematic trait screening preceded the green revolution’s yield gains by decades, and the seaweed sector is effectively at the beginning of that trajectory. Most cultivated seaweed genera have undergone little deliberate genetic improvement, meaning there may be substantial untapped potential even within currently farmed species once their traits are properly quantified and compared.

The framework also arrives at a moment when demand for seaweed products is diversifying beyond traditional food uses. Emerging applications, from biostimulants to alternative proteins, each favor different biochemical and physiological profiles, which strengthens the case for matching species to end-use through explicit trait data rather than trial and error. A species rich in specific polysaccharides, for example, may justify cultivation under conditions that would be uneconomical for a general-purpose food crop.

Finally, the perspective underscores a cultural shift within phycology itself. Much existing knowledge of seaweed biology resides in ecological studies of wild populations rather than in agronomic trials, and the proposed framework offers a route to translate that ecological understanding into cultivation insight. By treating traits as the common currency between wild ecology and farm performance, researchers can begin predicting how uncultivated species might behave in farm settings before expensive domestication programs begin, potentially shortening the path from ocean to aquaculture for promising new candidates.

Subject of Research: Using a trait-based framework to characterize farmed seaweeds and guide selection of new candidates

Article Title: Using a trait-based framework to characterize farmed seaweeds and guide selection of new candidates

Article References: Fong, C. R., Fong, P., & Froehlich, H. E. (2026). Using a trait-based framework to characterize farmed seaweeds and guide selection of new candidates. npj Ocean Sustainability. https://doi.org/10.1038/s44183-026-00239-5

Image Credits: AI Generated

DOI: 10.1038/s44183-026-00239-5

Keywords: trait-based, framework, characterize, farmed, seaweeds, guide, selection, candidates, scientific research

The heart of the new study is a technique called total infectome metatranscriptomics. Unlike targeted polymerase chain reaction assays, which can only detect organisms a researcher already suspects, RNA-based metatranscriptomic sequencing captures the full transcriptional activity of every microbe in a sample—RNA viruses, actively replicating DNA viruses, bacteria, fungi, and other eukaryotic microorganisms alike. Between 2021 and 2025, the team conducted epidemiological surveys across major grass carp-producing regions of China, including provinces in the Yangtze, Pearl, and Yellow River basins, confirming that OWS has spread well beyond its point of first documentation. From affected and healthy fish, the researchers dissected eight organs each—liver, spleen, kidney, intestine, gill, brain, muscle, and skin—and processed every organ as an independent sequencing library, generating 80 metatranscriptomes. After quality filtering and ribosomal RNA depletion, 8.09 billion high-quality reads remained, averaging 101.2 million per library, providing the depth needed for broad pathogen discovery.

The sequencing results revealed a strikingly complex microbial landscape. Across all libraries, the team identified 107 dominant microbial species: 32 viruses, 65 bacteria, and 10 eukaryotic microorganisms, spanning nine RNA viral supergroups, two DNA viral families, seven bacterial phyla, and seven eukaryotic phyla. Bacteria constituted the largest share of detected organisms at 60.7 percent, followed by RNA viruses at 22.4 percent, eukaryotes at 9.3 percent, and DNA viruses at 7.5 percent. Perhaps most tellingly, 68 of the 107 species—63.6 percent—were putatively novel, indicating that the majority of the grass carp-associated infectome had never been characterized before. Among the known and emerging agents were a grass carp hepacivirus, the first of its kind detected in this host; Chinook salmon nidovirus 1, previously reported only from salmonids; and a divergent aquareovirus the team named Shunde grass carp aquareovirus. Several parasitic eukaryotes from groups including Cnidaria, Euglenozoa, Fornicata, and Platyhelminthes also appeared in the dataset.

Detection alone, however, cannot establish causation—a lesson that has repeatedly frustrated disease investigators in aquaculture. To prioritize candidates, the researchers applied a comparative infectomics framework, quantifying microbial abundance as reads per million non-rRNA reads and retaining 33 taxa above a threshold of RPM ≥ 1. Differential abundance analysis, using a criterion of at least a fourfold change with a false discovery rate below 0.05, showed that overall microbial profiles clearly separated diseased fish from healthy controls. Among the enriched taxa, one organism stood out decisively: Flavobacterium psychrophilum was detected in every diseased individual, across multiple organs, and at its highest abundance in muscle and skin—precisely the tissues where OWS lesions were most severe. By contrast, parasitic eukaryotes such as Ichthyobodonidae, Trypanosomatidae, and Thelohanellus species showed inconsistent, sporadic occurrence, and the RNA viruses enriched in diseased fish phylogenetically clustered with invertebrate-associated lineages whose abundance correlated with parasite loads rather than direct infection of the fish.

With F. psychrophilum prioritized as the leading candidate, the team moved to experimental validation. The bacterium was isolated from lesion-associated muscle tissue of naturally diseased fish, yielding pale-yellow colonies on TYES agar after incubation at 15 degrees Celsius. Sequencing of the 16S rRNA gene placed the representative isolate, designated GC30-154, firmly within the F. psychrophilum clade with maximum bootstrap support. Healthy grass carp were then challenged by intramuscular injection with graded doses ranging from 10^4 to 10^8 colony-forming units. Control fish injected with buffer remained entirely healthy, while infected fish developed clinical signs beginning four days post-injection, with morbidity climbing in a dose-dependent fashion from 25 percent at the lowest dose to 100 percent at the highest. Mortality followed the same pattern, reaching 95 percent by day 18 in the highest-dose group, and the bacterium was successfully re-isolated from the lesions of deceased fish—satisfying key elements of Koch’s postulates.

The pathological picture in experimentally infected fish mirrored natural OWS with remarkable fidelity. Gross signs included focal erythema and swelling at the injection site, reddening around the pectoral-fin base, mild snout reddening, and tail erosion, while histopathology revealed severe muscle fiber degeneration, extensive vacuolation, and disruption of skin architecture—lesions closely resembling those in field cases, and concentrated in external and barrier tissues while liver, spleen, and kidney remained largely intact. Critically, the team then performed post-challenge total infectome analysis to rule out a role for secondary microbes in driving the experimental disease. Only F. psychrophilum appeared at consistently high abundance in infected animals, with the same muscle- and skin-dominant organ distribution seen in naturally diseased fish, while controls showed no signal whatsoever. The convergence of clinical signs, tissue pathology, mortality patterns, and infectome signatures established the bacterium as sufficient—and therefore the primary cause—of OWS.

A second layer of the investigation explained the disease’s peculiar seasonality. F. psychrophilum is classically regarded as a cold-water pathogen of salmonids, causing bacterial cold-water disease and rainbow trout fry syndrome, typically at temperatures below 10 degrees Celsius. Yet in grass carp the story was different. When challenged fish were held at constant temperatures of 10, 15, or 20 degrees Celsius, mortality was highest at 15 degrees—55 percent—with no deaths at the other temperatures during the observation period. More striking still was a temperature-shift experiment designed to mimic the overwintering-to-spring transition. Fish injected at 10 degrees and held there for 15 days showed no abnormalities; only when water temperature was gradually raised to 15 degrees did ulcers appear and mortality surge, reaching 95 percent within 14 days of warming. This thermal profile closely matches the late-overwintering and early-spring window in which natural OWS outbreaks occur, and it suggests the grass carp isolate may represent a host-adapted variant with altered temperature-dependent virulence.

Beyond the headline finding, the study carries broader implications for how infectious disease is investigated in complex systems. Aquatic environments teem with microbial diversity, and intensive aquaculture—shared water systems, high stocking densities, seasonal environmental stress—creates ideal conditions for polymicrobial communities to obscure etiology. The framework demonstrated here links epidemiological surveying, cohort-based comparative infectomics, targeted isolation, experimental infection, re-isolation, and post-challenge infectome validation into a coherent chain of evidence that converts unbiased pathogen discovery into causal inference. The authors emphasize that its success depends on careful attention to cohort representativeness, sampling coverage, and the detectability of pathogen-derived transcriptional signals, and that sampling and validation workflows must be tailored to the ecology of each disease system. Applied to OWS, the approach correctly demoted opportunistic eukaryotes and invertebrate-associated viruses that might otherwise have been mistaken for culprits, while flagging latent pathogen diversity—including novel hepaciviruses and nidoviruses—that could matter under future environmental or co-infection scenarios.

As aquaculture continues to expand and intensify worldwide, the connectivity between farming systems grows apace, raising the risk of pathogen transmission across previously separated host species and the emergence of new disease syndromes. The grass carp OWS resolution offers both a practical answer for producers—pointing toward surveillance and control of F. psychrophilum during spring warming—and a methodological template for wildlife, livestock, and even clinical medicine, where metagenomic detection increasingly outpaces causal interpretation. The study’s raw sequencing data have been deposited in a public aquatic pathogen platform, and all alignments and phylogenetic trees are openly available, reflecting the authors’ intent that the total infectome framework be adopted, adapted, and tested broadly. What began as an attempt to solve one stubborn disease in Chinese carp ponds may ultimately change how scientists everywhere distinguish the true cause of an outbreak from the microbial noise that surrounds it.

Subject of Research: A total infectome framework for resolving complex disease etiology in aquaculture

Article Title: A total infectome framework for resolving complex disease etiology in aquaculture

Article References: Xu, Y., Liu, H., Wu, W., Gu, Y., Zhang, N., Zhang, C., Zhou, R., Zhang, D., Weng, S., Shi, M., He, J., & He, J. (2026). A total infectome framework for resolving complex disease etiology in aquaculture. Advanced Biotechnology, 4(3), Article 31. https://doi.org/10.1007/s44307-026-00125-8

Image Credits: AI Generated

DOI: 10.1007/s44307-026-00125-8

Keywords: total, infectome, framework, resolving, complex, disease, etiology, aquaculture, scientific research

The METABRIC cohort, used for evaluation in this study, has become one of the most widely cited resources in breast cancer computational research. It was originally assembled through the Molecular Taxonomy of Breast Cancer International Consortium, which profiled tumor specimens with gene expression microarrays and single-nucleotide polymorphism arrays while collecting detailed clinical annotations, including survival time, vital status, tumor stage, grade, and treatment information. The availability of matched clinical, transcriptomic, and copy number data for the same patients makes it unusually well suited for multimodal modeling, since many cohorts provide only one molecular layer. Its public release through cBioPortal has enabled reproducible benchmarking, and numerous prognostic modeling studies have used it, which facilitates direct comparison of new methods against previously reported performance levels.

Copy number variation, one of the three data modalities integrated by the framework, captures gains and losses of chromosomal regions that occur frequently in breast tumors and can influence prognosis by altering gene dosage. Amplification of loci such as 17q12, which contains the ERBB2 gene, and loss of regions on chromosome 17p involving TP53 are well-characterized examples with direct clinical relevance. Unlike gene expression, which reflects dynamic transcriptional activity and is sensitive to sampling and processing conditions, copy number profiles are comparatively stable measurements of genomic alteration. Including this modality alongside transcriptomic and clinical data therefore supplies the model with information about the underlying genomic architecture of the tumor, complementing the more variable expression layer and the demographic and pathological detail in the clinical record.

The reported performance gap between the hybrid framework and the conventional stacking ensemble, 0.936 versus 0.898 ROC-AUC, is meaningful in the context of prognostic modeling, where incremental gains become progressively harder to achieve as models approach the ceiling imposed by noise in the outcome labels themselves. Survival endpoints in observational cohorts are affected by treatment heterogeneity, censoring, and variation in follow-up, all of which place an upper bound on achievable predictive accuracy. Gains of this magnitude, achieved while also improving recall and reducing false negatives, suggest that the stabilization provided by the bagging branch recovers predictive signal that a single stacking pass leaves buried in prediction variance rather than merely fitting noise more aggressively.

The emphasis on reducing false-negative predictions deserves particular attention from a clinical standpoint. In prognostic stratification, a false negative means a patient at genuinely elevated risk of poor outcome is classified as low risk and may be undertreated or monitored less intensively. The consequences of missing a high-risk patient are generally considered more severe than the consequences of flagging a low-risk patient for additional surveillance, which is why sensitivity and recall are weighted heavily in clinical risk model evaluation. The observation that the hybrid framework increased sensitivity relative to the stacking ensemble on the independent test set, without sacrificing overall discrimination, indicates that the improvement is concentrated where it matters most for patient management rather than distributed evenly across easy and difficult cases.

The use of stratified tenfold cross-validation alongside an independent test set reflects a methodological practice that strengthens confidence in the reported results. Cross-validation with stratification preserves the class balance of the outcome variable in each fold, which is important because survival outcomes in breast cancer cohorts are often imbalanced, with fewer events than censored observations. Evaluating across ten folds provides an estimate of how much performance fluctuates when the training composition changes, and the reported consistency in mean ROC-AUC, F1-score, balanced accuracy, and the Matthews correlation coefficient across folds suggests that the framework’s advantage is not an artifact of a single favorable data split. The Matthews correlation coefficient is especially informative in this setting because it summarizes all four cells of the confusion matrix and remains reliable under class imbalance, unlike raw accuracy.

The challenge of integrating heterogeneous data modalities is compounded by differences in dimensionality and scale across the layers involved. Clinical variables typically number in the dozens, while gene expression panels contribute thousands of features and copy number profiles span tens of thousands of genomic loci. Naive concatenation of such matrices allows the high-dimensional molecular layers to dominate the learned representation, potentially drowning out the compact but highly informative clinical signals such as tumor stage and nodal status. Meta-learning within a stacking architecture offers a partial solution, because the meta-learner operates on base model outputs rather than raw features, effectively giving each modality a chance to be distilled into a prediction before integration and reducing the risk that one data layer overwhelms the others.

The overfitting risk inherent in high-dimensional, low-sample-size settings, often described as the large p, small n problem, is a persistent obstacle in omics research. When the number of features approaches or exceeds the number of patients, models can achieve perfect fits to training data by exploiting spurious correlations that do not generalize. This is a central reason why many published prognostic models perform well in internal validation but fail in external cohorts. The design choice of using bagging to stabilize the meta-learning stage directly targets this failure mode, since averaging over bootstrap resamples dampens the influence of any particular subset of patients or features on the final prediction, producing decision boundaries that are less sensitive to sampling idiosyncrasies.

The broader trend toward multimodal artificial intelligence in oncology, noted in the study’s positioning of its contribution, has been driven largely by imaging applications, where deep learning on histopathology slides and radiological images has attracted the most investment and clinical validation effort. Structured multi-omics prognostic modeling has progressed more quietly, in part because the data are tabular and high-dimensional rather than spatially structured, making them less amenable to the convolutional and transformer architectures that have transformed image analysis. Ensemble methods built on tree-based and classical learners remain highly competitive for tabular biomedical data, and the finding that a carefully designed ensemble of such models outperforms more elaborate alternatives reinforces the point that architectural novelty is not always the limiting factor in predictive performance.

Weighted probability fusion, the final combination step in the framework, represents a simple but principled alternative to hard voting or unweighted averaging. By assigning different weights to the stacking and bagging branches, the framework can express the relative reliability of integrated multimodal inference versus stabilized prediction, and these weights can be tuned on validation data. Soft fusion over probability outputs also preserves more information than voting on discrete class labels, since the confidence of each branch contributes to the final decision. This granularity matters at the decision thresholds used clinically, where patients near the boundary between risk strata are precisely those for whom well-calibrated probabilities, rather than binary labels, are most useful.

From the perspective of clinical translation, several considerations follow from these results. Any prognostic model intended for prospective use must demonstrate generalization beyond the cohort on which it was developed, and while the independent test set used here provides an initial check, validation in external cohorts with different patient demographics, treatment protocols, and assay platforms remains the decisive test. The reliance on gene expression and copy number data also implies that deployment requires molecular profiling infrastructure, which is increasingly routine in oncology but still unevenly distributed. The stability of performance across validation folds, together with the open availability of the underlying cohort, positions this framework as a reproducible baseline against which future multimodal prognostic methods for structured omics data can be measured, and the demonstration that complementary ensemble roles improve robustness offers a design principle likely to extend to other cancer types where matched clinical and molecular data are available.

Subject of Research: A heterogeneous multimodal ensemble framework for multi-omics breast cancer prognosis

Article Title: A heterogeneous multimodal ensemble framework for multi-omics breast cancer prognosis

Article References: Bozorgpour, R., & Sadrabadi, M. S. (2026). A heterogeneous multimodal ensemble framework for multi-omics breast cancer prognosis. Clinical Cancer Bulletin, 5(1), Article 16. https://doi.org/10.1007/s44272-026-00068-0

Image Credits: AI Generated

DOI: 10.1007/s44272-026-00068-0

Keywords: heterogeneous, multimodal, ensemble, framework, multi-omics, breast, cancer, prognosis, scientific research