Chronic pain, defined as pain that recurs or persists for at least three months, affects between 12.7 and 33.7 percent of adults in Europe. People living with it often adapt their patterns of activity and rest to cope with both pain and daily demands, and these adaptations carry clinical weight. Fear of activity-related pain can lead to avoidance, while excessive overexertion produces a characteristic sawtooth pattern: individuals push through rising pain, then collapse into prolonged rest. Pacing, by contrast, is considered adaptive, involving the deliberate regulation of activity and rest. Because these patterns may reflect psychological state, physical function and pain interference, clinicians want objective ways to measure them. Yet in rehabilitation practice, activity patterns are still assessed mainly through questionnaires and interviews, which rely on recall and perception rather than actual behavior, and self-reported activity correlates only weakly with what activity monitors record.

Wrist-worn tri-axial accelerometers are user-friendly and yield valid and reliable data in the general population, but converting their raw signals into activity intensity classes has traditionally relied on cut-off point thresholds or regression-based approaches. Recent evidence suggests that deep learning models may outperform these conventional methods, offering high accuracy, adaptability across populations and the potential for continuous improvement as new data arrive. Two knowledge gaps, however, have limited their substantiated use in chronic pain care. First, the clinimetric properties of deep learning models have been investigated primarily in laboratory settings, leaving their construct validity under real-world conditions underexplored. Second, pain is known to alter motor behavior, from subtle stiffness and guarding to outright avoidance of painful movements, and these changes may distort accelerometer signals enough to reduce the accuracy of models trained on data from people without chronic pain. Until now, no study had investigated whether such models require adaptation to chronic pain-specific motor behavior.

To build the model, the researchers, led by Annet Doomen of HU University of Applied Sciences Utrecht and Maastricht University, assembled three datasets. Thirty healthy participants and thirty participants with chronic pain, recruited from interdisciplinary pain rehabilitation clinics in Eindhoven and Hoensbroek, each performed twenty-one predefined activities in a semi-controlled setting while wearing a tri-axial accelerometer sampling at 12.5 hertz on the lateral side of the dominant wrist. The activities, ranging from sitting reading a book to running and cycling, were labeled with metabolic equivalent of task values from the Adult Compendium of Physical Activities and classified into three intensity categories: sedentary, light and moderate-to-vigorous physical activity. Participants marked the start and end of each activity with three rapid forearm shakes, producing distinct acceleration peaks that allowed precise segmentation. The third dataset, Capture-24, is a large open-source collection of annotated real-world accelerometer data from 151 participants in Oxfordshire, United Kingdom, who wore an Axivity AX3 accelerometer and a neck-worn camera for approximately twenty-four hours in free-living conditions, with time-stamped camera images used to annotate the data.

After processing, the training data from 208 participants contained 211,864 ten-second windows, equivalent to 588 hours of labeled movement. The researchers tested four initial deep learning architectures and selected a convolutional neural network for its balance between performance and model size. They then trained this network with seven different strategies to evaluate performance across populations: training and testing on a single source, training on healthy data and validating on chronic pain data, and training on the combined datasets while validating separately on each. To address class imbalance, category-specific weights were applied during training, and the model was evaluated with repeated ten-fold cross-validation, with data split at the subject level so that each participant appeared exactly once in the validation set across the ten iterations. The Capture-24 data were downsampled from 100 to 12.5 hertz to match the laboratory data, a step that reduced accuracy when trained exclusively on Capture data but was more than compensated for by incorporating the laboratory data.

The results showed that classification accuracy reached 76.5 percent in individuals with chronic pain and 80.5 percent in individuals without chronic pain, with Macro-F1 scores of 73.4 and 78.3 percent respectively. The highest performance, 81.3 percent accuracy, was achieved when the model was trained, tested and validated exclusively on healthy participants, while strategies validated on the Capture data yielded the lowest scores. Notably, strategies incorporating the large Capture-24 dataset showed lower standard deviations and narrower performance ranges, suggesting that its inclusion enhanced the stability of the model’s output. Crucially, performance was comparable whether the model was trained on the pooled dataset, on healthy data alone, or exclusively on chronic pain data, indicating that chronic pain-specific training conferred no clear advantage within the populations studied. Sedentary intensity was correctly classified in 89 percent of chronic pain cases, light intensity in 64 percent and moderate-to-vigorous activity in 66 percent, with both light and moderate-to-vigorous activity more frequently underestimated than overestimated.

To explore construct validity in real-world settings, the researchers observed ten healthy participants and five individuals with chronic pain for approximately four hours each in their daily-life environments while they wore the accelerometer and followed their usual routines without modification. Observers recorded activity descriptions every minute, and two researchers independently assigned metabolic equivalent values to each recorded activity, achieving 81 percent inter-rater agreement before resolving all discrepancies. The deep learning model’s minute-by-minute predictions were then compared with these observations. Agreement was 78.6 percent in healthy participants and 73.6 percent in those with chronic pain, with weighted Cohen’s kappa values of 0.66 and 0.60 respectively, interpreted as substantial and moderate agreement. Individual agreement varied widely, ranging from 57.1 to 94.3 percent, and the lower agreement in chronic pain was primarily due to the model’s overclassification of sedentary activity in 26 percent of cases.

A closer inspection of the disagreements revealed multiple sources of error. Observer-based intensity classifications were sometimes inherently uncertain, particularly during minutes containing multiple activities with different metabolic demands, where assigning a single intensity category required averaging. The comparator instrument also relied on a predefined list of activities and corresponding metabolic equivalent values, irrespective of common individual variations in task execution; for example, the compendium assigns sitting and speaking a value of 1.3 without distinguishing between speaking with or without gesticulation, so when substantial arm movements occur during speech, the model may legitimately classify a higher intensity than the annotation suggests. More tellingly, several participants demonstrated systematic over- or underestimation across specific activities, indicating that individual differences in movement characteristics, rather than random classification errors, reduced construct validity. Post hoc analyses showed that two participants with chronic pain, diagnosed with fibromyalgia, patellofemoral pain syndrome, neuroma and non-specific low back pain, showed markedly lower performance metrics and substantially higher acceleration magnitudes during sedentary and light-intensity activities than the rest of the sample, suggesting elevated motor activity that the annotation procedure did not capture. Activities near the boundaries of the intensity categories posed additional challenges, as small differences around the metabolic cut-off values of 1.5 and 3.0 can result in different labels despite very similar movement patterns.

The authors acknowledge several limitations. Performance varied substantially between individuals, particularly within the chronic pain sample, so the overall results may have differed with a different group of participants. The awareness of being observed may have altered participants’ usual activities and movement patterns, and the five individuals with chronic pain included in the construct validity analysis reported lower pain intensity and less disability than those in the training data, meaning that individuals with more severe symptoms, who may display greater movement pattern deviations, were underrepresented. The study also focused specifically on activity intensities rather than the detection of pain-related movement behavior such as guarding, avoiding or rubbing. Before clinical or research application, the researchers recommend further investigation of construct validity, for example in individuals with chronic pain who exhibit pronounced differences in motor behavior, and of factors such as hypervigilance, obesity, pain location or the use of walking aids that may influence individual model performance. They also suggest exploring a continuous measure of activity intensity rather than discrete categories, which may help address borderline classification errors and enhance sensitivity to clinically meaningful change.

Nevertheless, the study represents a significant step toward standardizing activity intensity measurement in chronic pain research. All four initial deep learning models outperformed the conventional cut-points approach, consistent with previous findings in the general population, and the accompanying scripts and datasets enable accurate processing of raw accelerometer time series into activity intensities in the chronic pain populations studied. By making everything openly available on Zenodo, the team has lowered the barrier for other researchers to apply, scrutinize and improve the model. The work also delivers a valuable insight for the growing field of digital health: the inter-individual variability in agreement between real-world observations and model predictions highlights that the value of further investigating construct validity within a larger and more diverse sample of people with chronic pain is not just warranted but essential, if wearable-based activity monitoring is ever to deliver on its promise of truly personalized care.

Subject of Research: An open-source deep learning model for classifying real-world activity intensities in persons with chronic pain using a wrist-worn accelerometer

Article Title: An open-source deep learning model for classifying real-world activity intensities in persons with chronic pain using a wrist-worn accelerometer

Article References: Doomen, A., Felius, R., Wittink, H., Huijnen, I., Punt, M., & Smeets, R. (2026). An open-source deep learning model for classifying real-world activity intensities in persons with chronic pain using a wrist-worn accelerometer. Frontiers in Digital Health, 8, Article 1783507. https://doi.org/10.3389/fdgth.2026.1783507

Image Credits: AI Generated

DOI: 10.3389/fdgth.2026.1783507

Keywords: open-source, deep, learning, model, classifying, real-world, activity, intensities, persons, chronic, pain, wrist-worn

The software, published in the journal SoftwareX and available under a BSD-3-Clause license on GitHub, tackles a long-standing tension in battery modelling. On one side are equivalent circuit models, or ECMs—the simple electrical representations that dominate both academic research and real-world battery management systems. They treat a battery as a network of voltage sources, resistors, and capacitors, which makes them computationally cheap enough to embed in the microcontrollers of a battery management system, or BMS, where they handle safety, charge control, and diagnosis of the state of charge and state of health. On the other side are detailed physicochemical models, such as the famous Newman-type models, which resolve the electrochemistry of both electrodes but at a computational cost that rules out long-term simulation and embedded use. LIBquiv, developed by Jonas A. Braun, David Schmider, Simon Schwab, Patricia O. Mmeka, and Wolfgang G. Bessler, is designed to bridge that gap.

The key innovation is what the developers call physics-informed equivalent circuit models. Instead of a generic resistor labelled R1, LIBquiv allows users to specify separate resistances for the positive and negative electrodes, each of which can be parameterized as a function of current, temperature, or lithium stoichiometry. This electrode-resolved structure means that a purely electrical model suddenly carries electrochemical knowledge: the model components are chosen and parameterized according to the underlying physicochemical processes, while the computational complexity remains low. The framework also supports a core-shell particle diffusion model, capturing the transport dynamics that occur inside battery particles, and a hysteresis model that reproduces the voltage memory effects typical of lithium iron phosphate chemistry.

Temperature, which has a strong impact on battery operation and aging, is handled through a zero-dimensional thermal equivalent circuit. The thermal submodel tracks cell temperature using a heat capacity and heat transfer resistance, and it computes both reversible heat generation—derived from the entropy of the electrode materials—and irreversible heat generation from ohmic losses in the resistive elements. This coupling of electrical and thermal behavior in a single lightweight model is what allows LIBquiv to reproduce the strongly nonlinear thermal signatures observed in real cells, such as the large-format 150 ampere-hour LFP prismatic cell from the manufacturer Eikto used as a first demonstration case.

From the user’s perspective, LIBquiv is deliberately simple. The object-oriented implementation lets researchers operate a virtual battery object much as they would operate a real cell on a test bench. One-line commands initiate constant-current, constant-voltage, or constant-power operation, complete with cut-off criteria expressed as logic strings such as a minimum voltage or a maximum temperature. The class tracks simulation time, so multiple operations can be chained sequentially—the final state of one simulation becomes the initial state of the next—allowing complex protocols, like the CCCV charge-rest-discharge-rest sequence demonstrated in the examples, to be assembled as easily as programming a battery cycler. The full thermo-electrical simulation of the 150 Ah Eikto cell runs in just 0.27 seconds on a laptop.

Mathematically, LIBquiv represents the battery as a modular system of coupled time-dependent nonlinear differential-algebraic equations, typically of low order, solved simultaneously by standard MATLAB solvers. Each submodel—electrical, diffusion, hysteresis, aging, thermal—can be individually activated through dedicated parameters, so the mathematical complexity of the model can be adapted to the simulation task at hand. The developers caution that this flexibility comes with a trade-off: activating additional submodels introduces additional state variables and parameters that must be determined experimentally, so model predictions ultimately depend on the quality of the underlying parameterization. Model complexity, they advise, should be selected according to the intended application and the available data.

Perhaps the most striking demonstration is battery aging. Because of the low computational effort, LIBquiv can simulate months or years of battery performance, and because the models are physics-informed, degradation can be described using the concept of degradation modes—loss of lithium inventory and loss of active material. The team developed, parameterized, and validated an aging model for a cylindrical 3.35 Ah nickel cobalt oxide cell from Panasonic. A 650-hour aging simulation took only 139 seconds, and the results did more than reproduce the macroscopic capacity loss and resistance increase: they revealed the underlying degradation modes, showing that the notorious nonlinear aging knee in these cells arises from the interaction between internal resistance growth and lithium plating. That kind of interpretability—understanding not just that a battery degrades but why—is exactly what purely electrical models cannot deliver.

LIBquiv also opens the door to a novel diagnostic approach known as voltage-controlled modelling. Conventional simulations take current as input and predict voltage; LIBquiv can do the reverse, using measured cell voltage as the independent variable and simulating the current response along with the internal state of charge. The team demonstrated this on a 20 Ah lithium-ion pouch cell from Enertech, using measured voltage from a dynamic driving-cycle profile as model input. The simulated current agreed closely with the experimental data, allowing the state of charge to be inferred from internal model states without requiring measured current at all. The full diagnosis over 18,000 data points took just 7.92 seconds. Because modern vehicles and devices already log voltage continuously, this formulation could lead to innovative algorithms for diagnosing capacity, resistance, and even the open-circuit voltage curve of batteries in the field.

The software landscape for equivalent circuit simulation is crowded, spanning open-source projects in Python, C++, and MATLAB as well as commercial environments such as MathWorks Simscape Battery, Siemens Simcenter Amesim, and GT-SUITE. But according to the developers, no existing lightweight time-domain implementation combines electrode-resolved physics-informed models, hysteresis, degradation-mode-based aging, thermal behavior, and voltage-controlled simulation within a unified and intuitive framework. LIBquiv’s distinguishing feature is therefore not a single new circuit topology but the integration of multiple diagnostic and physicochemical extensions in one modular, computationally efficient package. The code has been tested with MATLAB R2019a, R2023b, and R2025b, and a reproducible capsule is archived on Zenodo.

The impact of the release extends across theory and practice. On the theoretical side, LIBquiv enables high-fidelity modelling and understanding-based interpretation of simulation results while remaining fast enough for long-term aging studies and parameter sweeps. On the practical side, the split architecture—model scripts separated from simulation scripts—mirrors the typical workflow of computational battery research and allows models to be exchanged easily between studies. Fully parameterized models of commercial cells ship with the code, and all three demonstration cases, from the thermo-electrical behavior of a large LFP cell to lifetime prediction and state-of-charge diagnosis, are provided in the GitHub repository and can be reproduced directly. For a field racing to electrify transport and stabilize grids with billions of cells, a free tool that makes battery models more physical, more transparent, and more accessible may prove to be one of the quiet catalysts of the energy transition.

Subject of Research: LIBquiv: An open-source MATLAB class for time-domain simulation of physics-informed equivalent circuit models of lithium-ion batteries

Article Title: LIBquiv: An open-source MATLAB class for time-domain simulation of physics-informed equivalent circuit models of lithium-ion batteries

Article References: Braun, J. A., Schmider, D., Schwab, S., Mmeka, P. O., & Bessler, W. G. (2026). LIBquiv: An open-source MATLAB class for time-domain simulation of physics-informed equivalent circuit models of lithium-ion batteries. SoftwareX, 36, Article 103088. https://doi.org/10.1016/j.softx.2026.103088

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103088

Keywords: LIBquiv, open-source, MATLAB, class, time-domain, simulation, physics-informed, equivalent, circuit, models, lithium-ion, batteries

That gap is now being addressed by SpectroElectroChem Suite, an integrated open-source graphical application described in the journal SoftwareX by Achim Habekost. The software, currently at version 5.5.2 and distributed under the MIT license, is built in Python and organised around the native data structures of spectro-electrochemical and RRDE experiments. It reads plain CSV files exported from any spectrometer or potentiostat, so proprietary formats stop being a barrier, and it makes established analysis methods usable without a single line of programming.

The motivation behind the suite reflects two practical obstacles that stand between raw instrument exports and interpretable results. The first is pre-processing. Broad fluorescence and scattering backgrounds can completely mask weak Raman bands, so baseline correction, smoothing, and robust peak detection must be performed before any interpretation, and the chosen parameters must be recorded for reproducibility. The second is quantitative RRDE evaluation, which involves a chain of linked calculations including collection efficiency, apparent electron number, peroxide yield, and mass-transport and kinetic analysis. Done manually in spreadsheets, these calculations are slow, error-prone, and poorly documented.

Existing tools each solve only part of the problem. Powerful open-source Python libraries such as pybaselines for baseline estimation, SciPy for filtering and peak finding, and RamanSPy for scriptable Raman pipelines are aimed squarely at users who write code, and none of them treat the electrochemical dimension as a native axis. Instrument-vendor software, by contrast, is tied to specific hardware and closed formats, is rarely inspectable, and typically treats the electrochemical dimension as an afterthought. The result is that experimental electrochemists, students, and technical staff are left with manual spreadsheet work that is slow and hard to verify.

SpectroElectroChem Suite is a modular desktop application whose central launcher opens four independent modules: Raman Spectrum Analysis, Raman/SERS Voltammogram, Absorpto-/Fluorovoltammogram, and RRDE Analysis. All modules share common input conventions and export routines. The numerical core relies on the scientific Python ecosystem, with NumPy and pandas for data handling, SciPy for Savitzky-Golay filtering and peak processing, pybaselines for baseline estimation, Matplotlib for static figures, Plotly for interactive HTML graphics, and OpenPyXL for Excel export. The graphical interfaces are built with PySide6 and Tkinter, and the suite is distributed both as a self-contained Windows executable requiring no separate Python installation and as a Python source distribution.

Two design principles run through every module. First, every processing stage can be inspected and exported simultaneously, including the raw, smoothed, and baseline-corrected values, the calculated baselines, and a metadata sheet recording the source file, the selected ranges, and all parameters. Second, any visualisation choice that alters the displayed magnitudes, such as percentile clipping, log1p scaling, or waterfall offsets, is recorded in the output so it can be reported explicitly rather than remaining hidden inside a figure. The three spectroscopic modules provide baseline correction using polynomial and penalised methods, Savitzky-Golay and moving-average smoothing, automatic peak detection with position, intensity, and full width at half maximum, tolerance-based identification against reference lists, and linked interactive three-dimensional surface, heatmap, contour, and rotatable waterfall views with numerical export.

The RRDE module goes further, offering a full quantitative analysis of rotating ring-disk experiments. After automatic import, which identifies the corresponding rotation speeds without manual column assignment, it provides ring-background correction options ranging from none to point-by-point subtraction of a nitrogen-atmosphere measurement, along with a user-defined or manufacturer-specified collection efficiency. For the oxygen reduction reaction, the software converts the cathodic disk current to its absolute value and enters the background-corrected ring current as an absolute value, a convention that prevents cancellation of current signs that can depend on potentiostat wiring and export conventions. It then calculates the apparent electron number and the peroxide yield, issuing a warning if the electron number falls outside the range of two to four or if the peroxide yield falls outside zero to one hundred percent. The module also calculates the reversible equilibrium potential from the selected reaction, pH, and temperature, and converts between reference-electrode scales so potentials can be expressed as overpotentials.

The numerical implementation covers the classical pillars of hydrodynamic electrochemistry. For Levich analysis, disk currents are linearly interpolated at user-selected potentials, the rotation rate is converted from revolutions per minute to angular velocity, and an unweighted least-squares regression of the limiting current against the square root of angular velocity yields the slope, intercept, and coefficient of determination. In a dedicated mode, the user supplies the electron number and the diffusion coefficient is calculated from the slope, or the diffusion coefficient is supplied and the electron number is calculated. The Koutecky-Levich relation regresses the reciprocal of the current magnitude against the inverse square root of angular velocity, with the intercept equal to the reciprocal of the kinetic current, requiring at least three rotation rates. Tafel analysis requires the user to select the linear kinetic interval manually in an interactive window, since the physically meaningful region depends on the reaction, and the software reports the slope in millivolts per decade, the intercept, the coefficient of determination, and the exchange current density obtained by extrapolation to zero overpotential. When mass-transport correction is enabled, the limiting current is estimated as the median current within a manually selected plateau range, with plateau scatter and drift reported as quality indicators. A Butler-Volmer estimate of the standard heterogeneous rate constant is available only for a well-defined one-step redox couple and is deliberately disabled for the oxygen reduction, oxygen evolution, hydrogen evolution, and hydrogen oxidation reactions, where unique oxidised and reduced bulk concentrations cannot be assigned.

The validation is notable for its transparency. The RRDE results were recalculated from the exported numerical data using separate linear regressions and the published equations, with a dataset comprising five rotation rates from 500 to 2,500 revolutions per minute, a collection efficiency of 0.25, a disk area of 0.196 square centimetres, a concentration of 1.2 times ten to the minus three moles per litre, and a kinematic viscosity of 0.010 square centimetres per second. Every independently recalculated quantity agreed with the software output within numerical precision. The peroxide yield of 25.9 percent and the electron number of 3.48 at a specific potential and rotation rate, the Levich slope and its coefficient of determination of 0.988, the diffusion coefficient of 9.20 times ten to the minus seven square centimetres per second, the Koutecky-Levich slope and its coefficient of determination of 0.978, the kinetic current of 1.02 times ten to the minus four amperes, the mean Tafel slope of 84.7 millivolts per decade, and the mean exchange current density of 1.46 times ten to the minus eleven amperes per square centimetre all matched exactly. In an illustrative alkaline oxygen reduction experiment on a platinum disk with a glassy carbon ring in oxygen-saturated 0.1 molar potassium hydroxide, the Levich regression at one selected potential gave a coefficient of determination of 0.9883 and the Koutecky-Levich regression at another gave 0.9778, with no out-of-range electron numbers or peroxide yields occurring within the evaluation range.

The impact of the suite lies less in a new algorithm than in an integrated, reproducible, GUI-based workflow spanning both spectroscopic and hydrodynamic electrochemistry. Because raw and processed values are preserved alongside the software version, its archived digital object identifier, and all parameters, any figure can be traced back to its underlying data, and an independent user can regenerate the same output using the archived CSV files and recorded parameters. The software was developed for and is used in university spectro-electrochemistry experiments, where transparent, verifiable processing steps have direct educational value by enabling students to see the effect of each pre-processing and evaluation choice. Development is public on GitHub with issue tracking, and each release is permanently archived on Zenodo under a citable identifier, supporting reuse and extension. The modular structure isolates the four workflows behind shared input and output code, allowing additional baseline models, further spectroscopic modalities, or batch processing to be added without disturbing existing modules, and future development will expand the automated test suite and continuous integration coverage.

Subject of Research: SpectroElectroChem Suite open-source software for spectro-electrochemical and RRDE data analysis

Article Title: SpectroElectroChem Suite: An integrated open-source graphical application for reproducible spectro-electrochemical and rotating ring–disk electrode data analysis

Article References: Habekost, A. (2026). SpectroElectroChem Suite: An integrated open-source graphical application for reproducible spectro-electrochemical and rotating ring–disk electrode data analysis. SoftwareX, 36, Article 103124. https://doi.org/10.1016/j.softx.2026.103124

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103124

Keywords: SpectroElectroChem Suite, spectro-electrochemistry, RRDE, oxygen reduction reaction, open-source software, reproducibility, SpectroElectroChem, Suite, integrated, open-source, graphical, application

The core insight is that when people search for a word online, the images they upload and tag collectively encode a consensus about what that word looks like. A search for “fire” returns images dominated by reds, oranges, and yellows; a search for “money” skews toward greens and grays. Rather than relying on small survey panels, the framework treats the entire indexed visual web as an enormous, self-selecting participant pool. The result is a color association profile for any word that a web image search engine supports, which in principle means nearly every major language on Earth.

Methodologically, the approach works by retrieving large numbers of images associated with a query word, extracting the dominant colors from each image, and aggregating those color distributions into a stable profile. Because the pipeline uses only openly available data and fully documented processing steps, anyone can rerun the analysis and obtain the same results. This stands in sharp contrast to recent computational attempts based on Large Language Models, which the authors argue suffer from a near-complete lack of transparency in both data and methodology. When a model’s internal reasoning cannot be inspected, its color associations cannot be independently verified, leading to low confidence, poor consistency, and unreliable outputs.

The survey-based tradition that dominated earlier work faced its own structural problems. Curating large word lists was laborious, enrolling enough participants to make results statistically meaningful was expensive, and researchers inevitably introduced bias when choosing which words and which color options to present. A participant forced to pick from a fixed palette may not perceive color the way the experimenters assumed. By contrast, the web-scale approach sidesteps these bottlenecks: the words come from the language itself, the images come from real-world usage, and the color extraction is performed by an explicit, auditable algorithm rather than by subjective human report.

To demonstrate the framework’s power, the researchers applied it to two very different domains: corporate branding and twentieth-century Anglo-American poetry. Branding is a natural test case, since companies spend enormous sums selecting colors that convey their identity. The analysis showed that the framework’s word-color associations align with patterns in actual color usage across corporate logos and visual identities, suggesting that the statistical consensus captured from web images reflects the same intuitions that professional designers exploit. For marketers, this offers a data-driven way to test whether a proposed brand palette genuinely matches the associations a name evokes in the public mind.

The poetry application is more surprising and arguably more interesting. Twentieth-century Anglo-American poetry is full of color words deployed for emotional and symbolic effect, from the bruised purples of despair to the bright yellows of joy. By mapping the words in poems to their statistically derived color associations, the researchers could analyze how poets actually used color language across the century, revealing patterns in color usage that literary scholars have long discussed qualitatively but never quantified at this scale. The framework thus provides a bridge between computational analysis and one of the most subjective corners of human expression.

What makes the study notable for a general audience is its accessibility. The code is freely available in open source, and the underlying image data comes from search engines that anyone can query. A high school student with a laptop could, in principle, generate a color profile for their own name, their favorite band, or an abstract concept like “freedom.” This democratization stands in stark contrast to proprietary AI systems whose color judgments arrive as opaque pronouncements. Transparency here is not just an academic virtue; it is what allows the results to be trusted, corrected, and extended by others.

The multilingual dimension may prove to be the framework’s most consequential feature. Because it relies on web image search rather than English-language text corpora, it can be readily adopted for any language supported by such engines. Color associations in Korean, Spanish, Arabic, or Hindi can be computed with the same pipeline, opening the door to comparative studies of whether color symbolism is universal or culturally specific. Do speakers of different languages associate “death” with the same colors? Does “luck” look red in Chinese-language image searches and green in Irish ones? The framework provides a standardized tool to ask these questions empirically rather than anecdotally.

The researchers emphasize that their goal is a universal, multilingual understanding of word-color association, and they expect the method to contribute new answers to long-standing questions in communication. The relationship between linguistic expression and visual cognition is one of the deepest puzzles in cognitive science, and having a scalable, unbiased measurement instrument changes what kinds of questions can be posed. Instead of debating whether a particular metaphor carries a color connotation, scholars can measure it. Instead of guessing which colors resonate with consumers, brands can check the statistical record.

There are, of course, limits worth keeping in mind. Web image data reflects the biases of the people who upload and tag images, and popular words will always yield richer data than obscure ones. The authors are candid that their contribution is a methodology, not a finished atlas of human color cognition. But that is precisely the point: by making the entire pipeline transparent and reproducible, they have converted a question that previously required expensive surveys and unverifiable AI outputs into one that anyone with an internet connection can investigate. In a field where intuition has long outrun evidence, that shift may be the most colorful result of all.

Subject of Research: Quantitative mapping of word-color associations using large-scale web image data

Article Title: Word-color association from large-scale online data: Methodology and applications

Article References: Wang, S., & Park, J. M. (2026). Word-color association from large-scale online data: Methodology and applications. PLOS One, 21(10), e0360407. https://doi.org/10.1371/journal.pone.0360407

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0360407

Keywords: word-color association, web image data, color psychology, large language models, corporate branding, poetry analysis, cognitive science, multilingual methodology, open source, PLOS ONE, data transparency, computational linguistics

At the heart of FuzzyDiary lies a decades-old mathematical idea known as fuzzy linguistic summarization, originally proposed by Ronald Yager in 1982. In this framework, statements take a protoform such as Q of the objects in R are S, where Q is a fuzzy quantifier like most, R a contextual qualifier such as in the afternoon, and S a linguistic term like low. Each instantiated statement carries a degree of truth, and in extended frameworks also measures of focus and simplicity, so the narrative is not merely plausible prose but a graded, quantifiable description of the data. Because these statements are produced deterministically, the same input signal and the same configuration always yield the same text, a property the developers consider non-negotiable in clinical settings where factual reliability cannot be compromised.

This determinism is what separates FuzzyDiary from the wave of generative approaches built on large language models. While LLMs can produce fluent summaries of monitoring data, the verbalization step may introduce unverified claims that cannot be traced back to the underlying observations. FuzzyDiary takes the opposite route: the domain knowledge is modeled explicitly in a human-readable YAML configuration, and the narrative layer resolves only the presentation of facts that protoform evaluation has already established. The authors argue that a cohesive report does not require a language model when the semantics are fully specified, and they position the verified linguistic facts produced by their framework as grounded input that downstream language models can consume rather than compete with.

Technically, FuzzyDiary organizes the analysis as a pipeline of six core modules plus a reporting layer, each with a well-defined input and output contract that allows users to invoke the full chain or any intermediate stage in isolation. The io module ingests univariate series from CSV, JSON, or in-memory Pandas DataFrames, standardizes them into a canonical representation indexed by calendar day, and handles resampling, imputation of missing values, and segmentation into daily windows. The unit of description is deliberately the calendar day, matching the natural reporting rhythm of clinical practice and keeping the narrative aligned with how clinicians, patients, and caregivers reason about longitudinal monitoring data.

The next stage applies the Ramer-Douglas-Peucker simplification algorithm, a geometric technique borrowed from cartography, to reduce each daily window to a compact piecewise-linear approximation. The tolerance parameter epsilon controls how closely the approximation follows the signal, and a built-in heuristic suggests a sensible default from the observed variance. Over this simplified substrate, the events module detects morphological events such as peaks, valleys, and prolonged episodes, which are declared declaratively in the configuration rather than hard-coded in the source. Each event is parameterized by the linguistic terms it spans, its minimum and maximum duration, and the gaps that may be merged, and events can be anchored to or composed from previously detected ones, enabling domain-specific patterns such as recovery from low, a return to normal values after a drop.

The describe module then evaluates protoform templates against the detected events and aggregated daily statistics, producing ranked statements with truth degrees. Because the terms of a fuzzy linguistic variable overlap by design, a single sample may hold membership in more than one term, and several statements can describe the same stretch of signal without anything in the output expressing how they relate. FuzzyDiary separates what is said from how it is said through a synthesis stage that orders summarizers and quantifiers by the centroids of their membership functions, attaches more specific statements below the general ones that cover them, and nests detected events under others that specialize them and contain them in time. The synthesis reorganizes the evidence but never discards it, keeping the complete list of instantiated protoforms with their degrees of truth in the report for verification.

A generation stage then renders the synthesized hierarchy as a cohesive paragraph through document planning, aggregation, lexicalization, and template-based surface realization. Document planning visits contextual scopes in the order declared in the configuration, making the narrative chronological, and assigns each scope a rhetorical relation such as rise, fall, or continuation by comparing the rank of its main summarizer with that of the preceding scope. Aggregation merges adjacent scopes described by the same quantifier and summarizer, while elision removes what the sentence has already said, so sibling statements sharing the quantifier of the main one are coordinated rather than repeated. Every generated clause retains the statements and events it summarizes, so the narrative is as traceable as the graded list beneath it.

The final artifact is a self-contained interactive HTML report in which the plotting library is embedded directly in the generated file, requiring no server and no network access at opening time. Hovering over a sentence reveals the statements and events it was built from, and clicking it highlights the corresponding segments on the signal panel, zooming the view to frame that interval. Clauses resting on day-level statements alone are linked to the temporal support of their contextual qualifier, so no sentence of the report is left without a visual counterpart. The developers demonstrated the pipeline on real continuous glucose monitoring data from the T1GDUJA dataset, collected with the approval of the Ethics Committee of the University of Jaén, with the linguistic variable, event catalog, and protoform templates defined together with clinical experts in diabetes care.

The authors see impact along several dimensions. By exposing morphological event detection and fuzzy protoform evaluation as composable modules driven by a single YAML configuration validated through a Pydantic schema, FuzzyDiary lowers the barrier to empirical work on linguistic descriptions of time series, making previously ad-hoc research questions tractable through configuration-only experiments. For practitioners in clinical reporting, ambient monitoring, industrial telemetry, or any longitudinal sensor stream, it offers a deterministic, traceable alternative to bespoke reporting scripts. Although clinical monitoring is the motivating use case, the framework is domain-agnostic by design: any univariate signal with a meaningful daily structure is handled by the same modules, and moving to a new domain requires only a new configuration, not a change to the source code.

Future releases will extend FuzzyDiary along two axes: support for multivariate signals through joint protoform templates, and a richer discourse layer that extends the rhetorical relations and aggregation rules to the comparison of several reporting periods. Integration with large language models is deliberately left outside the framework, since embedding one would trade the determinism and reproducibility of the output for variation in phrasing alone. The current version targets univariate signals at daily granularity, and the developers note that a structured evaluation of the usability of the reports with clinical professionals is left for future work. For now, the framework fills a concrete gap between tabular fuzzy linguistic summarization libraries and numerical time-series toolkits, occupying a previously uncovered niche in the open-source ecosystem for interpretable clinical data analysis.

Subject of Research: FuzzyDiary open-source framework for traceable linguistic summarization of clinical time series

Article Title: FuzzyDiary: An open-source framework for morphology-aware and traceable linguistic summarization of clinical time series

Article References: Gaitán-Guerrero, J. F., López-Ruiz, J. L., Martínez-Cruz, C., & Espinilla, M. (2026). FuzzyDiary: An open-source framework for morphology-aware and traceable linguistic summarization of clinical time series. SoftwareX, 36, Article 103135. https://doi.org/10.1016/j.softx.2026.103135

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103135

Keywords: FuzzyDiary, linguistic summarization, clinical time series, fuzzy logic, traceability, open-source software, explainable AI, natural language generation, open-source, framework, morphology-aware, traceable

The feature space is the framework that turns each data sample into a vector a model can analyze, and the authors argue that the success of any machine learning pipeline depends heavily on how well important features are selected and generated. Feature generation, as a central technique, converts raw data into a refined feature space that facilitates model training and enhances its structure. Yet despite steady progress in automated feature selection and generation, two major challenges persist: explainability and cross-domain generalization. Methods that work brilliantly on one dataset often fail to transfer to another, and the features they produce can be opaque to the experts who must act on them.

One popular alternative has been to let deep learning handle feature engineering automatically rather than relying on hand-crafted techniques. Deep learning models have proven highly effective across a variety of fields, but they are often considered less effective when applied to tabular datasets, where sample sizes are modest, columns are heterogeneous, and the signal is spread across interactions that neural architectures capture inefficiently. The İzmir team therefore took a different route: instead of asking a giant network to internalize the data, they asked small language models to reach outside the dataset entirely and pull in knowledge from the web.

The methodology centers on retrieval-augmented generation, the technique that lets a language model consult external sources before answering. For each tabular dataset, the system formulates search queries derived from the column names, values and the classification task itself. A small language model—compact enough to run without massive infrastructure—then processes the retrieved web content and distills it into concrete, machine-readable features. These web-search-based features are appended to the original tabular columns, enriching each row with external context that no amount of resampling the raw data could have produced.

The choice of a small language model rather than a frontier-scale one is deliberate. Small models are cheaper to run, easier to deploy, and—crucially for the explainability problem—their prompts and retrieval steps can be inspected. The study also emphasizes prompt design as a first-class component of the pipeline: how the model is asked to read a web page and translate it into a feature determines whether the extracted signal is meaningful or noise. Because the resulting features are grounded in retrieved documents rather than in the latent whims of a neural network, a domain expert can, in principle, trace why a model believes a certain external fact is relevant.

Experimental results show the approach reaching highest F1 scores of 0.383 and 0.848 on two well-known tabular datasets for the machine learning multi-class classification task. The F1 score, the harmonic mean of precision and recall, is a demanding metric for multi-class problems, where a model must balance accuracy across every category rather than excelling on the majority class. The gap between the two scores reflects the reality of tabular classification: some datasets carry rich external context that web knowledge can illuminate, while others resist enrichment because their structure is idiosyncratic or their semantics are not well represented online.

The broader significance lies in what this says about the evolving role of language models in data science. Retrieval-augmented generation has mostly been framed as a question-answering technology—a way to make chatbots cite their sources. Inan and Oguz repurpose it as a feature factory. The web becomes a vast, continuously updated auxiliary dataset, and the small language model becomes the extraction engine that converts unstructured human knowledge into the structured columns that classical classifiers consume. This reframing matters because it suggests that the value of language models for tabular problems may lie not in replacing gradient-boosted trees, but in feeding them better inputs.

It also speaks to the open-source momentum in the field. The study’s keywords—feature selection, retrieval-augmented generation, small language models, open-source and prompt design—map onto a research agenda that many laboratories can pursue without access to proprietary giants. A team with a modest GPU and a careful prompt library can replicate the pipeline, adapt the search strategies to their domain, and audit every feature the system proposes. That transparency directly addresses the explainability gap the authors identify as a central challenge of automated feature engineering.

Cross-domain generalization remains the harder test. A feature generated from web content is only as good as the match between the dataset’s semantics and the public knowledge the search engine surfaces. Medical datasets may benefit from clinical literature, while proprietary industrial sensor logs may find little useful context online. The reported F1 scores, while promising on the stronger dataset, are an honest reminder that enriching tabular data with web knowledge is not a universal accelerant. The authors’ contribution is a methodology and a demonstration, not a guarantee—and they present it as an alternative to deep-learning-based automatic feature engineering rather than a replacement for domain expertise.

Still, the vision is arresting. Imagine a near future in which every spreadsheet arrives pre-enriched: a language model has already searched the web for what each column means, retrieved the relevant facts, and generated features that encode them. Classification models would train on data that carries not just measurements but context, and analysts would see, in plain retrieved documents, why each new column exists. Inan and Oguz have taken an early, concrete step toward that future, showing that small language models can select and generate web-search-based features for tabular classification—and that the humble spreadsheet may finally have found its companion in the age of language models.

Subject of Research: Feature selection and generation with small language models and web search for tabular data classification

Article Title: Selecting and generating web search based features using small language models for tabular data classification

Article References: Inan, E., & Oguz, D. (2026). Selecting and generating web search based features using small language models for tabular data classification. Scientific Reports. https://doi.org/10.1038/s41598-026-75169-w

Image Credits: AI Generated

DOI: 10.1038/s41598-026-75169-w

Keywords: Feature selection, Retrieval-augmented generation, Small language models, Tabular data, Machine learning, Open-source, Prompt design, Selecting, generating, search, based, features

The project, named Fragmented Legacy: Turning Broken 3D Materials into Archaeological Interpretations, is one of 19 selected for this year’s Humanities and AI Virtual Institute awards from Schmidt Sciences. It is led by Chen Feng, Institute Associate Professor at NYU’s Tandon School of Engineering and founding co-director of the NYU Center for Robotics and Embodied Intelligence, together with Jing Zhang, a faculty fellow at the same center. On the archaeological side, the team includes Radu Iovita and Scott Williams of NYU’s Department of Anthropology and Raiko Krauß, Professor of Central and Southeast European Prehistory at Tübingen, with additional collaborators at Leiden University, Arizona State University, and the Max Planck Institute for Evolutionary Anthropology.

The problem the group is attacking is deceptively simple to state and brutally hard to solve. Most archaeological collections are never systematically refitted at all, because the work is so slow that it simply does not get done. Reported success rates tell the story: roughly 1 percent for ceramics, and between 4 and 26 percent for bones. A human specialist may spend days or weeks trying to find the matching edge for a single flake, and in a museum setting the physical constraints make it worse. Drawers can only be opened one at a time, so an entire collection can never be laid out for comparison at once. The result is that vast stores of information sit locked inside unmatched fragments.

Why does a single match matter so much? Because when two fragments fit, they carry information that neither piece holds alone. If a stone flake found in one part of a site fits back onto a core recovered from another, researchers can reconstruct how people moved through that space, how a tool was manufactured step by step, or whether the site’s layers were disturbed after deposition. Refitted animal bones can show how a carcass was butchered and how meat was shared across a group, a direct window into social behavior tens of thousands of years old. Refitted human fossils can reveal anatomical features that no single fragment preserves. Without refits, many of those questions simply go unanswered.

Feng’s lab specializes in teaching machines to understand three-dimensional shapes in the real world, where conditions are messy and incomplete. Broken archaeological material, he notes, is among the hardest versions of that problem: the pieces are worn, many are missing entirely, and there is no picture on the box showing what the finished object should look like. The AI systems his team is building must reason through that uncertainty, generating plausible reconstructions while flagging where the evidence is thin, with the archaeologists on the team judging whether the answers actually make sense in light of the archaeological record.

The technical plan unfolds in stages. First, the team will assemble a dataset of at least 100,000 real and digitally simulated broken objects, pairing 3D models with photographs, descriptions, and catalog data. Crucially, the simulated breaks will not be random: they will follow how stone, clay, and bone actually fracture, so that the synthetic fragments behave like the real ones archaeologists pull from the ground. Next, the researchers will develop AI models that work out where each fragment belongs while simultaneously generating a plausible version of the complete object. The models will reason across shape, images, and text, so that archaeological knowledge — typologies, raw material properties, contextual records — can rule out reconstructions that are geometrically possible but archaeologically implausible.

Archaeologists will remain in the loop throughout, comparing competing reconstructions and teaching the system which ones hold up. This human-in-the-loop design reflects a core philosophy of the project: the AI is not meant to replace expert judgment but to multiply its reach. As Zhang puts it, archaeologists are like Sherlock Holmes, piecing together clues to understand the past, and the goal is for Fragmented Legacy to be their Watson — matching fragments and suggesting possible reconstructions so that researchers have more time to interpret the evidence and ask new questions.

The project did not appear out of nowhere. It grew out of a collaboration between Feng’s AI4CE Lab and Iovita’s lab that began about four years ago with LUWA, an earlier system that used AI to classify microscopic wear on stone tools. Zhang, then a postdoctoral researcher working in both labs, connected the two groups. In 2024, she asked Iovita what archaeology’s most pressing technical problem was, and he pointed straight at refitting. That conversation produced GARF, a system that reassembles broken objects, and then CRAG, which goes further by also filling in missing pieces — work that laid the direct groundwork for the HAVI award now funding Fragmented Legacy.

Iovita’s ambitions for the technology extend well beyond reassembling objects within a single site. Refitting objects between sites, to demonstrate direct relationships between different locations, has only ever been done twice in the history of the discipline, he notes. His stated dream is to use AI to do this at scale, which he argues will revolutionize the field. If fragments from one site can be matched against fragments from another, researchers could trace the movement of raw materials, tools, and even the people who carried them across landscapes in ways that have previously been impractical or impossible.

Accessibility is built into the project from the start. All datasets, model weights, and benchmarks will be released under permissive open-source licenses, so that university researchers and cultural resource management firms — the companies that carry out much of the archaeological work required ahead of construction in the United States and abroad — can use the tools directly in their own workflows. The team also expects the methods to travel beyond archaeology, finding applications in other scientific fields and in practical repair and restoration work, wherever broken three-dimensional objects need to be reassembled and their missing parts intelligently reconstructed. For a discipline built on fragments, the promise is that the pieces sitting unmatched in museum drawers may finally begin to speak.

Subject of Research: AI-based refitting of fragmented archaeological artifacts such as stone tools, pottery, and bones

Article Title: NYU researchers are building AI to piece together broken archaeological finds

Article References: NYU researchers are building AI to piece together broken archaeological finds. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, archaeology, refitting, 3D modeling, stone tools, pottery, bones, NYU, Schmidt Sciences, machine learning, cultural heritage, open source

The core insight behind BGVD-State is that the integrity of review state matters as much as the discovery of new candidate issues. In a typical AI-assisted review, one model may observe a suspicious pattern, another may attach evidence from a scanner, and a third may render a verdict. If the recipient of a handoff only receives a transcript or a summary, they cannot reliably tell which candidate owns which piece of evidence, whether that evidence is still current, which investigation routes have already failed, or which verification decision controls whether the finding can be reported. Stale, incomplete, or contradictory evidence can slip through, or worse, a candidate’s lifecycle can be silently corrupted across handoffs.

BGVD-State, whose historical name stands for Blackboard-Guided Vulnerability Discovery, draws on an idea that predates modern AI by decades: the blackboard architecture, in which independent knowledge sources collaborate by reading and writing to a shared problem-solving state. The Hearsay-II speech understanding system of 1980 popularized the approach, and recent LLM research has revisited blackboards for multi-agent collaboration. What the new software adds is an explicit evidence chain and a reporting gate. Crucially, BGVD-State is not itself a language model. It is deterministic middleware that sits between event producers, such as agents, tool wrappers, parsers, and human analysts, and state consumers, such as stronger models, reviewers, and reporting pipelines.

The architecture works through typed event records. Every tool call, observation, evidence update, invalidation, or review decision is appended to an event store with a unique identifier, preserving submission order. From these ordered records, the EvidenceLifecycle component derives candidate status, evidence binding, verifier currency, and failed paths. A FinalizationGate then decides whether a candidate may enter a report, returning an allow or reject decision with machine-readable reasons. The whole system enforces five invariants: evidence stays bound to its candidate until explicitly invalidated; a candidate requires current material evidence and a current positive verifier result before reporting; the newest verifier result always governs; rejected routes are carried into later handoff packages so successors do not repeat failed paths; and finalization excludes invalidated or superseded evidence.

Because these rules are enforced by deterministic code rather than by prompts, model forgetting, reinterpretation, or prompt-rule violations cannot alter execution after context compression or a provider change. This is a deliberate design choice with significant implications. When the software version, policy configuration, and event sequence are identical, every client obtains the same lifecycle state and gate decision. Replay validates each event’s structure before appending it, and an event rejected by validation leaves the state unchanged. The middleware does not, however, verify the substantive truth of client-supplied observations; that responsibility remains with the producers and verifiers.

The authors demonstrate the system on a third-party review of fixed public revisions of real open-source projects, including TypeScript, Django, and an elliptic-curve cryptography library. A workflow of 22 event records reconstructed six suspected-risk candidates. Five candidates were rejected along paths whose exclusion reasons are fully recorded within those events. The remaining candidate had a locally reproduced technical observation and an initial positive technical review, but a subsequent security-impact review concluded that the available evidence and tests did not establish a privilege or trust-boundary impact sufficient for reporting. That negative decision was appended as a 23rd event, and after replay, BGVD-State generated a handoff package containing the current states, evidence, and review results for all six candidates, with the complete decision trail preserved for audit.

Reproducibility is treated as a first-class concern. Version 1.2.0, released under the MIT license, runs on Python 3.10 through 3.12 with no runtime dependencies, and no-install reviewer kits are available for Windows x64 and macOS on both Apple Silicon and Intel. The regression suite of 130 tests covers legacy-input normalization, rejected-event state snapshots, and artifact-completeness checks, none of which require an API key or a model call. In a historical performance profile cycling through six fixed synthetic candidates across scales from 100 to 100,000 events, each scale produced a single stable state hash and handoff hash across five repetitions, with median replay time rising from under a millisecond to about 1.34 seconds at the largest scale.

The paper is notably candid about where the approach helps and where it does not. In historical memory-representation studies, a schema-guided state format outperformed matched free-form memory on selected high-pressure continuation fixtures, with a valid-output sensitivity analysis showing six wins, no losses, and 23 ties. But under reduced pressure, with information-equivalent prose, and in a cross-family source-audit setting, the advantage disappeared or reversed, and the authors explicitly present the source-audit result as a tested negative boundary. A token-cost comparison using GLM 5.2 and DeepSeek v4-pro found that a route in which local weaker models handle routine state curation consumed roughly 24,000 strong-model finalizer tokens with no strong-model update tokens, versus nearly 147,000 total strong-model tokens for strong-model curation, with both routes reaching correct final decisions in all 30 runs.

The practical vision is one of composable infrastructure. Orchestration frameworks such as LangGraph can keep workflow checkpoints, and multi-agent systems like AutoGen can coordinate agents and memory, while BGVD-State retains sole authority over security-review state and the reporting gate, presenting the same event, handoff, and gate interfaces to whichever client is attached. Teams can swap models, prompts, retrievers, or tool wrappers without changing the rules that determine which evidence remains current and whether a finding may be reported. A weaker local model can maintain routine state while stronger models or human reviewers receive the current package only when needed.

The authors are careful about the limits of their claims. The applied case used local copies of public repository revisions without interacting with live services, and no candidate is presented as a maintainer-confirmed vulnerability. The middleware implements one deterministic policy family, external clients remain responsible for authenticating event sources, and deployment-level gains in discovery rate, review time, or total cost are not established by the reported checks. Future work points toward multi-writer transactions, cryptographic event integrity, and signed handoff exchange for distributed deployment. Even so, BGVD-State offers something the fast-moving world of AI security agents has lacked: a provider-independent, replayable contract for carrying evidence and state across handoffs, ensuring that when an AI says a vulnerability is worth reporting, the entire chain of evidence behind that claim can be traced, replayed, and trusted.

Subject of Research: Replayable middleware for evidence-gated state handoffs in third-party software security review using AI agents

Article Title: BGVD-State: Replayable middleware for evidence-gated handoffs in third-party software security review

Article References: Lv, Y., & Wang, C. (2026). BGVD-State: Replayable middleware for evidence-gated handoffs in third-party software security review. SoftwareX, 36, Article 103109. https://doi.org/10.1016/j.softx.2026.103109

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103109

Keywords: BGVD-State, software security, LLM agents, vulnerability discovery, blackboard architecture, middleware, reproducibility, evidence chain, handoff, open source, cybersecurity, multi-agent systems

The classic in question is the Aho-Corasick algorithm, a cornerstone of computer science first published in 1975. Its genius lies in building a finite-state automaton from an entire set of search patterns at once, so that a single left-to-right pass over the text can locate every occurrence of every pattern simultaneously. For exact matching, it remains remarkably hard to beat: no preprocessing of the text is required, the automaton is built once from the patterns, and the scan proceeds at a pace independent of how many patterns are in play. That property, known as index-free operation, is precisely what makes it attractive for proteomics workflows, where the pattern set, the peptides, changes with every experiment while the text, the protein database, may be reused.

The trouble begins with wildcards. Protein databases routinely contain the ambiguity codes B, Z and X, which stand for ‘asparagine or aspartic acid’, ‘glutamine or glutamic acid’ and ‘any amino acid’, respectively. When a peptide pattern collides with one of these characters in the text, a standard Aho-Corasick automaton simply has no transition to follow, and the match is lost. Existing solutions have been unsatisfying in different ways. FM-indexes, the compressed full-text indexes popularized by genomics, can handle wildcards but demand substantial preprocessing of the text and become awkward when the wildcard sits in the searched text rather than in the pattern. Other multi-pattern methods, such as the widely used Wu-Manber algorithm, handle exact matching quickly but offer no native wildcard semantics at all.

Wild-AC’s central idea is disarmingly elegant: when the scan encounters a wildcard character in the text, it does not halt or fall back to a slow secondary routine. Instead, it branches the search into parallel ‘scout’ paths, one for each possible interpretation of the ambiguous residue, while the unmodified primary search continues unaffected through the common, wildcard-free case. Each scout carries its own state within the automaton and explores the subtree of possibilities that the wildcard opens up. Because wildcards are relatively rare, typically a few percent of database positions after standard masking, the scouts remain short-lived excursions rather than an exponential explosion. The primary automaton, meanwhile, behaves exactly like a textbook Aho-Corasick machine, so the overwhelming majority of matching work pays no wildcard tax whatsoever.

The engineering details matter as much as the concept. Wild-AC is implemented in C++ and released as open source, with support for multi-threading so that large peptide sets can be partitioned across cores. Because no index of the text is built, the algorithm starts working immediately on any database in plain sequence format, a meaningful advantage in laboratory pipelines where databases are swapped frequently, for example when searching against species-specific proteomes or contamination databases. The memory footprint stays modest, and the automaton construction scales gracefully with the number of patterns, which in practice ranges from a thousand peptides in a targeted experiment to half a million in a deep discovery run.

The benchmark results are where the algorithm earns its headline. Bielow tested Wild-AC across realistic proteomics workloads: pattern sets of 1,000 to 500,000 peptides with an average length of roughly 18 amino acids, searched against protein databases ranging from 3 million to 210 million characters. In the wildcard case, Wild-AC outperformed the FM-index outright. In pure exact matching, it matched or exceeded Wu-Manber, which is notable because Wu-Manber was designed specifically for fast multi-pattern exact search and has few peers. The advantage held across the entire realistic range of pattern counts, suggesting the algorithm is not a niche performer tuned to one benchmark configuration but a genuine workhorse.

One nuance deserves attention. When the databases were artificially masked to a wildcard rate of 5 percent, a scenario representing heavily annotated or deliberately ambiguous reference proteomes, the picture became more conditional. Wild-AC retained its advantage for large peptide sets, but the FM-index became the preferable choice for small ones. This makes intuitive sense: an index amortizes its construction cost over many queries, so if you plan to run only a handful of small searches against the same text, paying once for a prebuilt index can win. For the typical proteomics pattern of use, many peptides per search and databases that change often, Wild-AC’s index-free design tips the balance decisively the other way.

Why does this matter beyond the benchmark tables? Peptide-to-protein mapping sits at the foundation of nearly every downstream analysis in shotgun proteomics: protein identification, quantification, quality control and the detection of sequence variants all depend on it being fast and correct. Ambiguous amino acids are not an exotic edge case. They appear wherever sequences are incompletely characterized, in genomes assembled from noisy data, in databases that merge paralogous proteins, and in deliberately degenerate searches for modified or variant peptides. An algorithm that treats wildcards as a first-class citizen, without imposing index construction or sacrificing exact-search speed, removes a persistent friction from those pipelines. The author credits discussions with Sandro Andreotti on the algorithmic design and implementation input from Simon Gene Gottlieb, whose FM-index codebase and fuzzy amino acid matching implementation served as the benchmark comparison, with reviewer feedback from Ragnar Groot Koerkamp among others helping sharpen the final manuscript.

There is also a broader lesson for computational biology in how Wild-AC was built. Rather than inventing an entirely new data structure, the work takes a mature, well-understood algorithm and extends it minimally to cover a missing capability. The scout-path mechanism preserves the automaton’s behavior in the common case and pays for wildcard handling only where it is actually needed. This kind of surgical extension, validating performance against the strongest existing tools rather than straw men, is exactly the pattern that tends to produce software that gets adopted. The open-source C++ implementation, hosted publicly on GitHub, lowers the barrier for integration into existing search engines and quality-control tools, where Aho-Corasick machinery is often already present in some form.

For the proteomics community, the immediate takeaway is practical: if your pipeline maps large peptide sets against protein databases containing ambiguity codes, Wild-AC now offers the fastest known route, with no index to maintain and threads to spare. For small searches against a fixed, heavily masked database, the FM-index remains a sensible choice. For everyone else, the paper is a reminder that some of the most impactful bioinformatics advances still come from revisiting the classics of stringology and asking what happens when the text, not just the pattern, refuses to be unambiguous. As protein databases continue to swell with sequences of uneven certainty, algorithms that embrace that uncertainty at full speed will only grow in importance.

Subject of Research: A wildcard-enabled multi-pattern string matching algorithm for peptide-to-protein mapping in proteomics

Article Title: Wild-AC: A fast, index-free multi-pattern string matching algorithm with wildcard support for proteomics

Article References: Bielow, C. (2026). Wild-AC: A fast, index-free multi-pattern string matching algorithm with wildcard support for proteomics. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06686-8

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06686-8

Keywords: Wild-AC, Aho-Corasick, string matching, wildcards, proteomics, peptide mapping, protein databases, FM-index, Wu-Manber, algorithms, bioinformatics, open source

A new study published in Nature Methods by Jacob Schreiber, of the Research Institute of Molecular Pathology at the Vienna BioCenter and UMass Chan Medical School, aims to close that gap. The paper introduces tangermeme, a free, open-source Python package described as an everything-but-the-model toolkit for genomic deep learning. The name of the design philosophy is deliberate. Rather than competing with the many frameworks that build, train and host neural networks, tangermeme concentrates exclusively on what happens after a model has been trained: making predictions efficiently, perturbing sequences, attributing importance to individual nucleotides, calling regulatory patterns and designing new DNA. The result, according to the paper, is a comprehensive, flexible and efficient Swiss Army knife for cis-regulatory pattern identification and analysis.

The rationale for this division of labor is grounded in how the field has actually evolved. Model architectures and training strategies churn constantly, with convolutions giving way to transformers and new optimizers arriving every year. Downstream analyses, by contrast, are remarkably stable and largely agnostic to architecture. Estimating the effect of a noncoding variant on a model’s prediction works essentially the same way whether the underlying network uses convolutional filters or self-attention, and whether it was trained with one optimizer or another. This modularity means that a well-engineered library of analysis operations can serve any model, old or new. Yet until now, no optimized repository of such methods existed that generalized across model types. Instead, researchers typically ship bespoke analysis code bundled with each released model, and existing packages either implement only a few related algorithms or focus on training and fine-tuning with limited post-training support.

Tangermeme’s central technical trick is the strict separation of sequence manipulations from model operations, which can then be freely stacked to create analyses. In silico marginalization compares predictions before and after substituting a short sequence into a template, quickly identifying which motifs from a database the model responds to. Ablation is the conceptual opposite, altering a region the model may be using and measuring the drop in prediction. Variant effect estimation compares predictions before and after one or a few noncontiguous substitutions, allowing fine-mapping of candidate regulatory variants. Crucially, the operation performed after a sequence manipulation need not be a simple prediction; it can be DeepLIFT/SHAP attribution, in silico saturation mutagenesis, or any custom operation a researcher writes. This composability lets scientists ask questions such as how surrounding sequence context influences transcription factor binding, whether a regulatory element is suitable for a synthetic construct, or how two variants at the same locus interact, all in a handful of commands.

The package also embraces DNA design, implementing several methods catalogued in a recent review of enhancer modeling. The simplest is screening: generate random sequences, score them against a user-defined objective, and keep the best. Greedy substitution, sometimes called directed evolution, iteratively improves a sequence by testing every possible single-nucleotide change and keeping the best one at each step. Motif implantation works similarly but swaps in entire motifs from a database rather than single bases. Most novel is a construct marginalization method that merges design with marginalization, creating short DNA inserts that reliably shift model predictions, such as patterns of predicted chromatin accessibility, averaged across many background sequences, without requiring prior knowledge of what the construct should contain. Such libraries of designed inserts could prove valuable for synthetic biology and therapeutic applications.

Engineering choices under the hood are where tangermeme distinguishes itself from prior efforts. Operations come with built-in batching, support every data type and device available in PyTorch, and work out of the box on models with multiple inputs and outputs, arbitrary internal units including convolutions, long short-term memory blocks and transformers, and any output format from a single binary value to a full base-pair-resolution profile. Speed benchmarks illustrate the payoff: tangermeme one-hot encodes the entirety of human chromosome 1 in under two seconds, roughly three times faster than the next fastest implementation tested. That may sound trivial, but fast encoding enables on-the-fly batch generation during analysis, and the same optimization philosophy runs throughout the codebase. Batching also solves a practical bottleneck in DeepLIFT/SHAP, which in many implementations requires all background sequences to fit in GPU memory simultaneously, a serious constraint for massive models.

Perhaps the most consequential finding to emerge from building the toolkit carefully is a subtle failure mode in widely used attribution code. DeepLIFT/SHAP works by overriding the backward pass of nonlinear operations to incorporate a reference sequence, and identifying which operations are nonlinear requires checking each computational layer against an internal dictionary. Existing implementations can silently fail when they encounter a nonlinear operation outside that dictionary or when an activation is reused, often without documentation or warnings. The telltale sign is convergence deltas, which should be zero in theory and near machine precision in practice, ballooning to indicate that the attributions are simply wrong. Tangermeme never fails silently: it raises warnings when convergence deltas are too high, can return those deltas for monitoring, and allows users to register custom nonlinear functions to avoid the problem entirely.

Beyond wrapping established algorithms, tangermeme introduces genuinely new methods for distilling learned cis-regulatory logic, centered on objects called seqlets, the contiguous spans of high-attribution nucleotides that attribution methods such as DeepLIFT/SHAP, in silico saturation mutagenesis, PISA or saliency produce. The package’s novel recursive seqlet caller is, according to the paper, the first to call variable-length seqlets directly, using a principled statistical definition: a seqlet is a span whose attribution sum is statistically significant against a null distribution, with the requirement that all subspans within it also pass the test. The algorithm bins attribution values, builds null distributions for each candidate length, converts them to P values, and decodes the longest non-overlapping significant spans. Compared against a reimplementation of TF-MoDISco’s caller on synthetic sequences with an inserted MYC motif, the recursive approach produced shorter, more precise calls and ran faster on modest-to-large numbers of sequences, while TF-MoDISco’s wider, more permissive calls reflect design choices made for its original role inside a larger pipeline.

The payoff of this machinery is demonstrated on real models. Applying the automatic seqlet-calling and annotation pipeline, which maps called seqlets to motifs from the JASPAR database and then counts motif occurrences, pairwise co-occurrences and spacing relationships, Schreiber compared two models ostensibly trained to predict the same thing: MYC binding. At the PLD6 promoter, attribution analysis revealed that BPNet’s predictions are driven almost exclusively by a MYC motif, whereas the much larger Beluga model relies on a broader lexicon of motifs also associated with chromatin accessibility and transcription initiation. Counting annotated seqlets across all MYC peaks and their pairwise occurrences made these differences in learned regulatory logic immediately visible, despite both models achieving similar predictive goals.

Tangermeme is released under the MIT license, installable with a single pip command and accompanied by extensive documentation and tutorials that reproduce the paper’s figures. Although the demonstrations focus on transcription factor binding and chromatin accessibility, the authors emphasize that the toolkit applies to models of any genomic modality, including alternative splicing, transcription, enhancer activity, RNA stability and RNA binding. Schreiber also points toward the future: as large language model-based coding agents become common in research, well-documented packages with tested, composable building blocks will let such systems reuse reliable code rather than reinventing analyses, reducing the burden of auditing generated code. Future development will focus on expanding functionality and deepening documentation for exactly that purpose, positioning tangermeme as infrastructure for an era in which interpreting the genome’s regulatory code may depend as much on trustworthy software as on the models themselves.

Subject of Research: A deep learning interpretation toolkit for cis-regulatory genomics

Article Title: Tangermeme: a toolkit for understanding cis-regulatory logic using deep learning models

Article References: Schreiber, J. (2026). Tangermeme: a toolkit for understanding cis-regulatory logic using deep learning models. Nature Methods. https://doi.org/10.1038/s41592-026-03254-z

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03254-z

Keywords: deep learning, genomics, cis-regulatory logic, software toolkit, feature attribution, seqlets, transcription factor binding, variant effect prediction, DNA design, interpretability, Nature Methods, open source