The focus of this investigation was the semicore electrons in alkali metals, a group renowned for their exceptional reactivity and position in the first column of the periodic table. Using sophisticated quantum chemical simulations powered by high-performance computational facilities, the team explored how these electrons influence phase transitions in compounds formed between alkali metals and fluorine. Their work was motivated by the notorious B1-B2 structural phase transition, a pressure-induced rearrangement of atomic crystals from an octahedral to a cubic lattice structure, familiar from the classical sodium chloride to cesium chloride conversion.

Quantum chemical calculations harnessed to unravel this complexity rely on approximations designed to make the famously intractable Schrödinger equation solvable for systems involving many interacting electrons. Historically, conventional wisdom attributed the need for immense pressures—on the order of hundreds of gigapascals—to activate core electron bond participation. This study defies that notion by demonstrating electron bonding activation at pressure levels far less severe, within the range of a few gigapascals, a regime found not only deep within Earth’s crust but intriguingly close to everyday atmospheric pressures.

In a stunning discovery, the researchers uncovered that cesium, among the heaviest alkali metals, exhibits semicore electron bonding even under ambient conditions. This defies previous theories that core electron involvement was exclusive to extreme planetary interiors. By analyzing cesium chloride’s crystal structure, the team deduced that the B2 phase—characterized by a cubic lattice stabilized by semicore electron activity—exists naturally without the necessity of high-pressure environments.

The implications of such a finding extend beyond academic curiosity. If semicore electrons contribute to bonding under conditions previously considered benign, this necessitates a reappraisal of theoretical models that predict elemental behavior in the Earth’s mantle and cores of terrestrial planets. The participation of these electrons could fundamentally influence key geophysical phenomena such as a planet’s density profile, tectonic dynamics, and magnetic field generation.

The team, led by SUNY Distinguished Professor Eva Zurek and co-researcher Stefano Racioppi, utilized state-of-the-art computational models facilitated by the University at Buffalo’s Center for Computational Research. Their modeling offers a high-resolution window into electron density distributions and bonding interactions that had eluded experimental characterization thus far. This theoretical advance outlines a paradigm shift in the understanding of chemical elements under pressure, with semicore electrons playing an indispensable role previously underestimated or overlooked in chemical physics.

Even more compelling is how the study’s insights could ripple into planetary science. If electrons alter their bonding behavior as a function of pressure in ways that differ from established expectations, current models of planetary formation and evolution could be incomplete or inaccurate. For example, shifts in bonding states at lower pressures could impact material properties that control mantle convection and core dynamics right down to magnetic field intensity and stability, both crucial for planetary habitability.

While these revelations stem primarily from computational simulation, the authors are cautious yet hopeful that experimental verification is within reach. They propose targeted experiments involving X-ray diffraction under controlled pressures to validate the role of semicore electrons in the bonding transformation and the B1-B2 phase transition. Such efforts could cement the theoretical predictions, potentially reshaping textbooks and inspiring new directions in materials chemistry and geophysics.

This investigation into semicore electron activation stands as a powerful reminder of how even well-established scientific doctrines remain open to challenge with the emergence of novel technology and rigorous inquiry. By pushing the boundaries of quantum chemical modeling, the researchers at Buffalo have not only illuminated the nuanced behavior of electrons deeply embedded in atomic structures but also opened pathways to understanding the underlying chemistry that defines planetary compositions and transformations.

With the support of the U.S. National Science Foundation’s Center for Matter at Atomic Pressure, this fusion of quantum computational chemistry and geophysical relevance represents the vanguard of interdisciplinary research. It underscores how foundational electron interactions—at scales far smaller than conventional chemical bonds—can dictate macroscopic properties that influence entire planetary bodies and conditions for life.

Ultimately, this study propels a revisitation of fundamental chemical bonding theories. It compels scientists to consider that electrons formerly deemed inert within atoms may emerge as active agents under a spectrum of hitherto unexpected pressures. This evolution in understanding expands the horizon for material science, planetary geology, and the quest to decode the complexities of matter both on Earth and across the cosmos.

Subject of Research:
Quantum chemical behavior of semicore electrons in alkali metals under pressure

Article Title:
Activation of Semicore Electrons in Alkali Metals and Their Role in the B1–B2 Phase Transition under Pressure

News Publication Date:
25-Aug-2025

Web References:
https://pubs.acs.org/doi/10.1021/jacs.5c08582

Image Credits:
Eva Zurek/University at Buffalo

Keywords:
Chemical elements, Chemical structure, Covalent bonds, Molecular chemistry, Materials, Heavy metals, Geophysics, Quantum mechanics

Dyslexia and dysgraphia are neurodevelopmental disorders that profoundly affect children’s learning. Dyslexia primarily impairs reading and language processing abilities, while dysgraphia manifests as difficulties with handwriting and fine motor skills. Early identification of these disorders is essential to mitigate their long-term impact on academic achievement and socio-emotional development. The team at the University at Buffalo, led by SUNY Distinguished Professor Venu Govindaraju in the Department of Computer Science and Engineering, is pioneering AI methodologies aimed at revolutionizing the screening process, especially in underserved communities where resources like speech-language pathologists and occupational therapists are scarce.

The project builds on decades of pioneering work by Govindaraju and his colleagues in the realm of handwriting recognition, which historically leveraged machine learning and natural language processing to automate mail sorting for the U.S. Postal Service. In this new iteration, the research extends AI’s capabilities to recognize nuanced handwriting patterns indicative of dyslexia and dysgraphia, such as irregular letter formation, inconsistent spacing, spelling errors, and disorganized writing structure. By deciphering these subtle cues from handwritten samples, the AI system offers a multifaceted approach that identifies both motor-based and cognitive markers of these disorders.

While prior research in this domain has concentrated more heavily on dysgraphia due to its discernible motor symptoms, the new study significantly amplifies focus on dyslexia’s more elusive signs. Dyslexia’s hallmark difficulties in language processing do not always prominently manifest in handwriting, complicating early detection efforts. Nevertheless, the research identifies specific behavioral indicators embedded within the act of writing—such as frequent spelling mistakes and letter reversals—that can serve as red flags when analyzed through sophisticated AI algorithms.

A notable challenge the researchers confronted involved the scarcity of handwriting samples available from children, especially those diagnosed with these learning disabilities, to effectively train AI models. To overcome this, the team collected a broad dataset consisting of both paper and digital handwriting samples from kindergarten through fifth-grade students at an elementary school in Reno, Nevada. This ethically approved and anonymized collection effort provided a rich foundation with which the AI system could be trained, validated, and refined to ensure accuracy and real-world applicability.

Integral to the development process was the collaboration with educators, speech-language pathologists, and occupational therapists. Their unique insights ensured the AI tools aligned with practical classroom environments and clinical evaluations. This end-user informed approach not only enhances the tool’s usability but also increases its potential for adoption across various educational and therapeutic settings.

The research further integrates the Dysgraphia and Dyslexia Behavioral Indicator Checklist (DDBIC), co-developed by literacy expert Dr. Abbie Olszewski from the University of Nevada, Reno. The DDBIC catalogues 17 behavioral cues observable before, during, and after writing, offering a standardized framework for symptom identification. The AI models are being trained to autonomously perform the DDBIC screening, streamlining what currently requires specialist evaluation and manual observation.

Central to the technology is a sophisticated suite of AI models tasked with analyzing multiple dimensions of handwriting. These include the detection of motor control difficulties through metrics such as writing speed, pen pressure, and stroke movements; examination of visual handwriting features like letter size, spacing, and slant; and conversion of handwriting to digitized text for linguistic analysis focusing on misspellings, letter reversals, and grammatical errors. Collectively, these models integrate to unearth cognitive as well as physical markers indicative of the disorders.

The culmination of this research is the development of a comprehensive AI assessment tool that synthesizes inputs from various models into a unified diagnostic summary. This holistic evaluation platform not only flags potential neurodevelopmental concerns but could also provide educators and clinicians with actionable insights to tailor early interventions, addressing a crucial bottleneck in early childhood education systems.

Beyond its technological sophistication, the study underscores the potential of AI for social good. By democratizing access to reliable screening tools, it aims to level the playing field for children in underserved and remote regions where trained specialists are often unavailable. Early intervention enabled by such AI tools could transform educational trajectories, preventing the compounding effects of untreated dyslexia and dysgraphia.

While this research is ongoing, its implications resonate widely. It is a rare example of applied AI synergizing with education and healthcare, showcasing how machine learning and natural language processing advancements can directly enhance human well-being. The interdisciplinary nature of this work, incorporating computer science, linguistics, education, and clinical practice, exemplifies the collaborative spirit needed to tackle complex neurodevelopmental challenges.

The initiative is part of the National AI Institute for Exceptional Education, a University at Buffalo-led research consortium focused on developing AI systems that identify and assist children with speech and language processing difficulties. Funding from the U.S. National Science Foundation supports this cutting-edge endeavor, lending critical resources to push the boundaries of AI applications in public health.

Co-authors contributing to this research include Bharat Jayarman, director at the Amrita Institute of Advanced Research and professor emeritus at UB; Srirangaraj Setlur, principal research scientist at the UB Center for Unified Biometrics and Sensors; and doctoral researcher Sahana Rangasrinivasan, who emphasizes the criticality of building AI tools from the standpoint of those who will employ them. Their collective expertise adds profound depth to the project’s interdisciplinary approach.

This latest advancement in AI-powered handwriting analysis marks a promising shift in detection methodology for dyslexia and dysgraphia, promising greater accessibility, speed, and accuracy in diagnosis. By harnessing the power of contemporary AI combined with behavioral science, the University at Buffalo team sets a high bar for innovation in educational technology and neurodevelopmental health, heralding a future where early intervention is not a privilege but a standard available to all children.

Subject of Research: Early Detection of Dyslexia and Dysgraphia Using Artificial Intelligence-Powered Handwriting Analysis

Article Title: University at Buffalo Develops AI-Based Handwriting Analysis Tool for Early Detection of Dyslexia and Dysgraphia in Children

News Publication Date: Not specified in provided text

Web References: DOI: 10.1007/s42979-025-03927-0 (Published in SN Computer Science)

References: Research article published in SN Computer Science; National AI Institute for Exceptional Education project details

Image Credits: Not provided

The USMLE, comprising three essential Step exams, assesses a physician’s mastery not only of factual knowledge but also fundamental patient-centered skills and clinical judgment. SCAI’s remarkable performance—achieving an accuracy score of 95.2% on Step 3—surpasses that of GPT-4 Omni’s score of 90.5%, placing it well ahead of native large language models operating without semantic augmentation. This indicates a qualitative leap rather than an incremental gain in AI-driven diagnostics. The development team, led by Dr. Peter L. Elkin, Chair of Biomedical Informatics at the Jacobs School of Medicine and UBMD Internal Medicine, emphasizes that SCAI is designed to serve as a collaborative partner for clinicians, not a replacement.

Unlike conventional AI tools that rely heavily on statistical associations gleaned from massive digital text corpora—tools occasionally criticized for effectively “plagiarizing” internet content—SCAI employs a rigorously structured approach to knowledge representation and reasoning. The system utilizes semantic triples, a distinct form of knowledge encoding that links subjects, relations, and objects (for example, “Penicillin treats pneumococcal pneumonia”) to establish a richly interconnected knowledge network. This semantic network forms the substrate from which SCAI can draw logical inferences, enabling reasoning that resembles the cognitive processes physicians develop over years of training.

The integration of formal semantics into large language models represents a paradigm shift that directly addresses a known limitation in AI medicine: confabulation. Confabulation refers to an AI’s tendency to produce plausible but unverified or erroneous responses when faced with insufficient data. By coupling semantic knowledge graphs—structures designed to discover both explicit and hidden relationships in heterogeneous medical datasets—with retrieval-augmented generation techniques, SCAI drastically reduces this issue. Retrieval-augmented generation allows the model to consult specific, authoritative external knowledge databases before formulating responses, effectively ensuring that its outputs are anchored in validated medical facts.

This architecture leverages approximately 13 million discrete medical facts, encompassing verified data from an extensive array of domains, including genomic information, clinical guidelines, drug interactions, patient safety recommendations, and discharge protocols. Notably, the system deliberately excludes potentially biased inputs, such as raw clinical notes, to preserve objectivity and evidence-based reasoning. This breadth and depth of curated information imbue SCAI with a unique capacity for comprehensive and nuanced analysis across specialties.

SCAI’s conversational ability is another critical feature that differentiates it from traditional AI assistants. It can engage interactively with clinicians and lay users alike, enhancing decision-making through reasoned dialogue rather than rote answer generation. This dynamic interaction simulates a partnership, where the AI contributes balanced, evidence-based perspectives, helping users to explore complex clinical scenarios more thoroughly. Dr. Elkin asserts that by embedding semantic understanding into AI models, they are nurturing systems that think more like trained physicians, who integrate evidence-based medicine into patient care with contextual reasoning.

Beyond improving diagnostic accuracy, SCAI has the potential to transform healthcare delivery at multiple levels. Its ability to democratize access to specialist knowledge can empower primary care providers to manage more complex conditions confidently. This could alleviate bottlenecks in specialty referrals, reduce disparities in access to expert care, and enhance patient safety by better informing clinical decisions with a comprehensive, real-time knowledge base. Furthermore, SCAI’s scalable platform could serve as an educational tool, assisting medical students and professionals in navigating the increasingly complex landscape of modern medicine.

Despite the system’s extraordinary capabilities, the research team unequivocally positions AI as an augmentation rather than a replacement for human clinicians. As Dr. Elkin eloquently puts it, “Artificial intelligence isn’t going to replace doctors, but a doctor who uses AI may replace a doctor who does not.” This underscores a future healthcare environment where AI tools serve as cognitive amplifiers, sharpening a physician’s clinical acumen and supporting more precise and informed patient care decisions.

The origins of SCAI trace back to an existing foundation in natural language processing, which the research team significantly advanced by enriching it with authoritative, multi-dimensional medical knowledge. This fusion required expert curation and computational ingenuity to construct semantic knowledge networks capable of reasoning. The result is an AI tool that conceptualizes medical facts semantically rather than statistically, enabling it to infer “causal” relationships and logical consequences in a manner far closer to human clinical thought.

In terms of methodological innovation, SCAI exemplifies the power of combining symbolic AI—knowledge graphs and semantic triples—with modern deep learning architectures. This hybrid approach leverages the strengths of both paradigms: symbolic AI’s interpretability and reasoning, alongside the pattern recognition and generative capabilities of neural networks. Retrieval-augmented generation acts as an intelligent interface, fetching exact information when needed, grounding the AI model in authentic medical knowledge rather than probabilistic guesswork.

The research team comprises a multidisciplinary roster of experts across biomedical informatics, oncology, and veterans’ healthcare systems, highlighting the collaborative effort that underpinned SCAI’s creation. Funded by the National Institutes of Health and the Department of Veterans Affairs, their work reflects a strategic investment in AI technologies poised to revolutionize clinical practice. The open-access publication allows for widespread dissemination and invites further validation and development within the broader scientific and medical communities.

As AI continues to evolve, SCAI represents a model for future clinical decision support systems—ones that do not merely compute but reason, engage, and integrate seamlessly into the human workflows of medicine. Its debut performance on the USMLE not only sets a new bar for technical achievement but also sparks a profound conversation on how AI can best serve the healthcare ecosystem. In an era where medicine demands both precision and empathy, tools like SCAI promise to be indispensable allies in the quest to improve patient outcomes worldwide.

Subject of Research:
Not applicable

Article Title:
Semantic Clinical Artificial Intelligence vs Native Large Language Model Performance on the USMLE

News Publication Date:
22-Apr-2025

Web References:
http://jamanetwork.com/journals/jamanetworkopen/fullarticle/10.1001/jamanetworkopen.2025.6359?utm_source=For_The_Media&utm_medium=referral&utm_campaign=ftm_links&utm_term=042225

References:
10.1001/jamanetworkopen.2025.6359

Image Credits:
Credit: Sandra Kicman/University at Buffalo

Keywords:
Informatics