Browse Articles

Prime editing enables drug-controllable T-cell therapies with clinical immunosuppression

Current cell and gene therapies lack clinically practical mechanisms to selectively promote or suppress therapeutic cells in vivo, a limitation that is particularly acute in patients requiring ongoing immunosuppression. This includes gene therapy for immune dysregulation syndromes, and antigen-specific or chimeric antigen receptor (CAR) T-cell therapy for patients requiring immunosuppression (e.g., transplant recipients), where both pathogenic and therapeutic cells may be suppressed. Here, we develop a multiplex prime-editing platform that converts commonly used immunosuppressive drugs into tools for in vivo control of T-cell therapies via defined, pathway-specific drug resistance. Focusing initially on gene therapy, prime editing efficiently edited loci of multiple pathogenic variants associated with immune dysregulation in primary human T-cells and corrected the HAVCR2 driver mutation in T-cells from multiple patients with subcutaneous panniculitis-like T-cell lymphoma (SPTCL). Comprehensive genomic, transcriptional, immunophenotypic, and clonal analyses demonstrated minimal off-target perturbation. Multiplexed gene correction and drug-resistance editing of T-cells from patients with SPTCL enabled selective in vivo expansion of corrected cells under immunosuppressive pressure in humanized mouse models and exhibited retained sensitivity to alternative agents permitting rapid in vivo suppression. Extending this approach, prime edited, drug-resistant antigen-specific and CAR T-cells retained effector function despite pharmacologic immunosuppression, demonstrating the generalizability of this platform to diverse cellular therapies. Together, these findings establish multiplex prime editing as a promising preclinical framework for generating drug-controllable T-cell therapies, enabling selective in vivo modulation in settings where immunosuppression cannot be withdrawn.

Adaptive level modification via player skill classification and large language models

Abstract Maintaining player engagement in video games requires a careful balance between challenge and player competence. Static difficulty settings fail to account for individual skill variation, while existing dynamic difficulty adjustment systems are limited to tuning low-level game parameters rather than restructuring level content. This paper presents an adaptive level modification framework that personalizes gameplay by continuously inferring player skill and applying targeted structural modifications to level content in real-time. A hybrid behavioral dataset is constructed by combining agent-generated trajectories, produced by Proximal Policy Optimization (PPO) agents, a reinforcement learning approach, trained at three distinct skill levels, with manually collected human gameplay data labeled through clustering. A classifier trained on this dataset categorizes players into expert, normal, and beginner skill levels, achieving an overall accuracy of 97.82%. The classifier output drives a two-stage large language model (LLM) pipeline guided by prompt engineering, which expands a skill-conditioned prompt into a structured modification instruction applied to the current level chunk. A physics-constrained verifier based on a graph-based shortest path method ensures all modified levels remain traversable. Evaluated on Super Mario Bros. levels, the framework achieves a post-modification playability rate of 74.1% at the full-level granularity and 83.5% at the isolated-chunk granularity, closely matching the 80.0% baseline of the original levels.

Power-law transient leakage in Czochralski silicon pn-junctions: Dispersive transport induced by exponential trap distributions

We propose an analytical model for the power-law transient leakage current observed in pn-junctions fabricated in Czochralski silicon (Cz-Si). The model assumes an exponential distribution of trap states within the bandgap, attributed to oxygen-related defects. The transient current is analyzed within the framework of the multiple-trapping model of dispersive transport, originally developed for disordered semiconductors. Carrier populations in the conduction band, shallow traps, and deeply frozen traps are evaluated using the density of states and Fermi–Dirac statistics. The power-law decay of the leakage current is derived as a drift-limited hole current in the neutral region, where carriers previously swept out from the junction are transported with an effective drift mobility governed by the evolving quasi-Fermi level within the exponential trap distribution. The model provides a unified physical interpretation of the observed power-law behavior, suggesting that the exponential trap distribution may originate from interfacial states, band-tail states, or border traps associated with oxygen aggregation.

Ligand regulation and function of preformed EGFR dimers

Receptor tyrosine kinases (RTKs) are key therapeutic targets in cancer, diabetes, and other diseases. With only one transmembrane α-helix—compared with seven in G-protein-coupled receptors—RTKs are thought to be activated by ligand-induced dimerization. Complicating this view, however, one of the best-studied RTKs, the insulin receptor (IR), forms allosterically regulated covalent dimers. Moreover, noncovalent “preformed” dimers have frequently been reported for the sequence-related epidermal growth factor receptor (EGFR), one of the first RTKs for which ligand-induced dimerization was described. Here, we describe a detailed structural view of a preformed EGFR dimer. Using cryo-EM, we describe how the Caenorhabditis elegans EGFR (LET-23) dimerizes without ligand. We show that preformed dimer formation modulates ligand sensitivity in vivo, but is not required for signaling itself. We also elucidate substantial ligand-induced conformational changes in LET-23 required for signaling. Our structures reveal unexpected similarities between regulation of LET-23 and the IR, suggesting that LET-23 may represent an evolutionary “missing link” between the IR and EGFR families. In the absence of ligand, intermolecular interactions within preformed receptor dimers hold the extracellular juxtamembrane regions far apart to separate the intracellular kinase domains so that they remain inactive. Ligand binding disrupts these interactions to remove the restraints on the kinase domains, which then can associate to become activated. Our analysis further suggests a unified model for the allosteric activation of preformed RTK dimers that has important implications for understanding cell-surface EGFR.

Human HSPCs clones balance stochastic diversification with cytokine-induced differentiation

Hematopoietic stem and progenitor cells (HSPCs) balance self-renewal with on-demand differentiation of blood lineages. Modern bone marrow atlases specify static transcriptional states, but not HSPC dynamics, which are only partly understood, particularly in humans. Here, we introduce cHSPCTrack (clonal HSPC tracker), an in vitro framework for serially tracking clonal differentiation and fate acquisition from human HSPCs extracted from circulation. By following thousands of single-cell-derived clones stimulated by different cytokine combinations, cHSPCTrack reconstructs the balance between stochasticity and directed differentiation in human HSPCs. Single HSPCs stochastically yield multiple lineages under a uniform cytokine environment, with stimulatory inputs modulating differentiation probabilities by introducing biases toward preferred fates. In parallel, differentiation rates vary stochastically across and within clones. Clonal memory stabilizes the induction of a few genes per clone over otherwise archetypical differentiation trajectories. Within clones, multiple fates can interact with each other, as shown for clones mixing basophils, eosinophils, and mast cells with other fates. cHSPCTrack opens avenues for understanding hematopoietic clonal dynamics in health and disease.

OZAIANet a fusion based explainable deep learning framework for thermal image based diabetes classification

Structural distortion and small-polaron conduction in Ga-doped LiMn2O4 cathode material

Ga-doped LiMn2O4 was synthesized via solid-state reaction. X-ray photoelectron spectroscopy and x-ray powder diffraction confirm that the samples crystallize in the Fd3¯m space group of the spinel structure, where Ga3+ partially replaces Mn3+ at 16d sites, thereby increasing the average Mn oxidation state. Rietveld refinements indicate that while the lattice parameter remains almost unchanged, the oxygen positional parameter u decreases toward 0.25 with increasing Ga content. Consequently, the M–O (M: metal ions at a 16d site) bond length and the O–M–O angle both increase. Ga substitution suppresses the trigonal distortion of the MO6 octahedron, bringing it closer to a regular octahedral geometry. Electrical conductivity and thermopower measurements demonstrate that hopping conduction in Ga-doped LiMn2O4 involves nonadiabatic small polarons, which are eg electrons on Mn3+ ions, and that Ga substitution decreases not only the carrier density but also the small-polaron hopping energy WH. The reduction of WH increases the polaron mobility and decreases the polaron mass, explaining the enhanced electrical conductivity for x ≤ 0.1 despite a reduced carrier density. Importantly, the suppression of trigonal distortion is considered to decrease the polaron binding energy WP, which arises from the interaction between charge carriers and the lattice distortion. Furthermore, the increased O–M–O angle, reflecting the suppression of trigonal distortion, is expected to enhance the hybridization between the Mn 3d and O 2p wave functions. This structural relaxation of the trigonal distortion appears to be more effective in reducing WH than the inhibition of the Jahn–Teller distortion.

There is no free benchmark: An institutional view of legal AI benchmarking

Despite substantial excitement around the use of AI in law, little information exists on the performance and associated risks of the domain’s widely marketed tools. Recent work, for instance, has demonstrated the significant potential for “hallucinations”—wherein models make up facts, law, and precedent—leading Chief Justice Roberts to spotlight this risk in his annual report on the judiciary. We argue that there is a need for public AI benchmarking in law. First, relative to other AI application domains, the legal AI ecosystem lacks legibility—there is little information about the design and performance of many commercial legal AI systems. Legal AI has not benefited from the types of benchmarking that have catalyzed, measured, and informed AI innovation and responsible use in other domains. Second, we articulate the challenges of the institutional design of benchmarking. We illustrate how benchmarks can be captured, watered down, and abused. Careful institutional design around the why, who, what, and how of benchmarking will be critical to navigate difficult tradeoffs of transparency, objectivity, expertise, and resources. Third, addressing legal AI’s illegibility requires matching institutional models to available resources and constraints. Rather than advocating for a single “best” approach to benchmarking, we show how benchmarking strategies depend on available resources.

Lisocabtagene maraleucel combined with ibrutinib in R/R CLL or SLL: primary results from TRANSCEND CLL 004

Patients in the liso-cel plus ibrutinib cohort of the phase 1/2, open-label TRANSCEND CLL 004 study had relapsed/refractory chronic lymphocytic leukemia (CLL)/small lymphocytic lymphoma (SLL) and received liso-cel (50×106 [dose level (DL)1] or 100×106 [DL2] chimeric antigen receptor-positive T cells) with concurrent ibrutinib from enrollment through 90 days after liso-cel infusion or longer per investigator discretion. Primary end point was complete response/remission (CR)/CR with incomplete marrow recovery (CRi) by investigator assessment. Among 56 patients who received ibrutinib plus liso-cel (DL1, n=5; DL2, n=51), median (range) age was 64.5 years (44‒77), 98% had high-risk cytogenetics, 55% had progression on Bruton tyrosine kinase inhibitor and venetoclax failure, and median (range) number of prior therapies was 5 (1‒13). Median (range) follow-up was 24.8 months (3.1‒51.8). At DL2, CR/CRi rate was 45% (95% confidence interval [CI], 31‒60) and overall response rate was 86% (95% CI, 74‒94). Median (95% CI) duration of response was not reached (NR; 28.7‒NR) for patients with CR/CRi and 41.4 months (23.3‒NR) for all responders. Median (95% CI) progression-free survival was 31.4 months (20.1‒NR). In DL1+DL2, most common grade ≥3 treatment-emergent adverse events (TEAE) were neutropenia (52%) and anemia (41%). Cytokine release syndrome was reported in 80% of patients (grade 3, 4%; no grade 4/5), and neurological events in 41% (grade 3/4, 11%; no grade 5). Ibrutinib-related TEAEs were reported in 68% of patients (grade 3/4, 43%; no grade 5). Liso-cel plus ibrutinib demonstrated notable efficacy in patients with relapsed/refractory CLL/SLL with predictable and manageable safety. Clinicaltrials.gov: NCT03331198; NCT03435796.

Association between thoracic hyperkyphosis and sagittal skeletal malocclusion patterns: a cross-sectional study

Water-assisted laser ablation of silicon carbide: Damage evolution and underlying mechanisms

To address severe thermal damage in silicon carbide laser processing, this study employs molecular dynamics (MD) to investigate atomistic damage evolution following ultrafast energy deposition in SiC under vacuum and water-assisted conditions. The findings demonstrate that the aqueous medium significantly suppresses the heat-affected zone, yielding an approximately 43% reduction in damaged atoms in the present MD model. Through coupled rapid quenching and interfacial confinement, the water layer attenuates stress redistribution, restricts damage near the surface, and helps preserve a locally ordered near-surface layer while a localized subsurface phase-transformation zone remains. Furthermore, Partial radial distribution function analysis indicates that the water-assisted condition better preserves local Si–C coordination and medium-range sublattice correlations within the simulated time window. Within the present MD energy-input window, the simulated crater morphology exhibits a transition from depth-dominated to width-dominated growth between 6000 and 6500 eV/ps, indicating that lateral energy redistribution becomes increasingly dominant under water-assisted conditions. This research elucidates how the aqueous medium regulates energy partition and damage evolution via a synergistic “thermal dissipation-mechanical confinement-stress modulation” mechanism, providing atomistic-level mechanistic guidance for interpreting experimentally observable damage suppression and morphology evolution in water-assisted ultrafast laser processing of SiC.

Measuring disparate impact in human and machine decisions

Empirical analyses have grown increasingly important in discrimination litigation with the greater availability of detailed data on individuals and decisions. A popular analytic strategy is to estimate disparities after adjusting for observed covariates, typically with a regression model, in hopes of ferreting out discriminatory intent. This approach, however, is ill-suited to auditing algorithms that are now commonly used to aid decisions, which typically do not include race or other legally protected factors as inputs. Motivated by legal understandings of disparate impact, we introduce an approach that aims to measure “unjustified” disparities in both human and machine decisions. Our method, which we call risk-adjusted regression, proceeds in three steps. In the first step, we combine all available information in a machine learning model to estimate the value, or inversely, the risk, of taking a certain action, such as approving a loan application or hiring a job candidate. Second, we measure disparities in decisions after adjusting for these risk estimates alone. Finally, in the third step, we assess the sensitivity of results to potential mismeasurement of risk. We demonstrate this approach on a detailed dataset of 2.2 million police stops of pedestrians in New York City, and show that traditional statistical tests of discrimination can substantially understate the magnitude of (risk-adjusted) racial disparities.

CD4+ T cells orchestrate the immune response to ALK-positive T-cell lymphoma

Immunotherapy has revolutionized the treatment of solid cancers in recent years. However, T-cell lymphomas (T-NHLs) originate from immune cells themselves and are biologically heterogeneous, rendering investigations of immune checkpoint inhibitor (ICI) mechanisms of action complex. While case reports and individual Anaplastic Large Cell Lymphoma (ALCL) cases enrolled in T-NHL trials demonstrated favourable responses to ICI, hyperprogression was observed in other T-NHL subtypes. We therefore utilized a syngeneic mouse model of ALK+ ALCL to investigate immune surveillance and ICI-induced immune response. Transplantation experiments combined with depletion of relevant immune axes revealed that ALCL immune surveillance is mediated by CD4+ T cells and NK cells. Innate and adaptive immune cell infiltration was confirmed on a large series of primary human ALK+ ALCL samples. ICI monotherapy demonstrated robust efficacy in murine ALCL, inducing complete remissions in approximately 50% of treated animals. Mechanistically, PD-L1 blockade reversed Treg-mediated immunosuppression and increased the frequency of circulating effector CD8+ T lymphocytes, thereby prolonging survival significantly. Importantly, CD4+ T cells proved indispensable for driving and sustaining immunotherapy-induced anti-tumour responses in murine ALCL. CD4+ T cells of non-responder animals exhibited an exhausted phenotype and a transcriptomic Th22-like signature, implicating persistent T-cell exhaustion and polarization as a meaningful immune-evasion mechanism. Our findings uncover CD4+ T cells as key players in spontaneous and immunotherapy-mediated anti T-cell lymphoma immunity, which demonstrates the critically needed preclinical proof-of-concept for the safe and effective use of immunotherapy for ALCL.

Comparative evaluation of six machine learning models for multi-fuel variable compression ratio diesel engine emission prediction under leave-one-out cross-validation

Ballistic contributions to mesoscale thermal conductivity: Closing a flux-based model with lattice–Boltzmann–Peierls–Callaway simulations

The thermal transport in dielectric and semiconductor solids departs from classical diffusion when characteristic dimensions approach phonon scattering lengths, producing mixed ballistic–diffusive behavior. Although effective conductivity is often correlated with the Knudsen number (the ratio of scattering length to sample size), such formulations do not uniquely capture phonon backscattering or provide a consistent definition for reformulating bulk properties outside the diffusive infinite sample size limit. In the scope of this work, (i) the McKelvey–Shockley Flux Method (MSFM) is extended to decompose heat flux into ballistic and diffusive contributions, and (ii) a mesoscale thermal law is derived to explicitly relate the effective thermal conductivity with ballistic thermal conductance, Knudsen number, effective backscattering length, and bulk-like thermal conductivity. To identify the latter two parameters, we perform systematic steady-state numerical experiments using a D3Q19 Lattice–Boltzmann–Peierls–Callaway (LBPC) model with separate relaxation channels for normal and Umklapp scattering. From LBPC solutions, we extract an effective phonon mean free path (equivalent to backscattering length) and a bulk-like thermal conductivity from thickness sweep regressions and use them to close the MSFM-based constitutive relations. These LBPC-extracted closure parameters are validated against existing theoretical modeling frameworks, as well as experimental data from the literature on a range of different materials. The findings reflect quantitative consistency between MSFM and LBPC as independent descriptors of ballistic–diffusive phonon transport, providing an opportunity to reduce the degrees of freedom in the model. This self-consistent mesoscale description links microscopic scattering to macroscopic behavior and supports incorporating ballistic effects into continuum models relevant to ultrafast heating, thermoelectrics, nanoelectronics, thermophones, and nanoscale information storage.

Continuity in geometric intuition between humans and monkeys

Geometric intuition is regarded as a hallmark of human intelligence. Humans hold a unique advantage in recognizing geometric shapes with Euclidean features (e.g., parallelism, symmetry) and have more sophisticated geometric concepts than nonhuman primates. A popular explanation for this human distinctiveness is that humans rely on discrete, symbolic features to encode geometric shapes, while other primates depend only on continuous perceptual features. The current study challenges this view. We used a match-to-sample task to test whether monkeys share geometric representations with humans. We found that monkeys combined high-level perceptual representations with abstract symbolic representations like humans, suggesting a graded, quantitative continuity in geometric intuition across species rather than a categorical, qualitative divide. Importantly, monkeys showed stronger reliance on symbolic features when processing rotated geometric shapes than preschoolers did with unrotated shapes, indicating that what appears to be symbolic representation may largely emerge from rotation-invariant encoding. These results challenge the hypothesis of human uniqueness in geometric intuition and suggest that its origins in humans are rooted in abstractions shared with primates.

OPTI-AML: Prospective Comparison of 28 vs.14 days of Venetoclax Induction with Azacitidine in Older Adults with AML

Azacitidine (Aza) plus venetoclax (Ven) is standard treatment for older/unfit patients with newly diagnosed (ND) acute myeloid leukemia (AML). The approved 28-day (D) Ven schedule is associated with prolonged cytopenias, causing frequent dose reductions and cycle delays. Retrospective studies show similar efficacy and reduced toxicity with abbreviated Ven dosing, but prospective data is lacking. We conducted OPTI-AML(NCT03013998), a prospective randomized phase 2 trial comparing 28D Ven (AV28) versus 14D (AV14) with Aza (75mg/m²x7D) for C1-2 in genomically agnostic ND-AML patients ≥60 years. The primary endpoint was complete remission (CR) rate achieved at any time with two cycles of therapy. Between 2023-2025, 169 patients received AV28 (n=83) or AV14 (n=86). CR across two cycles was 49.4% (AV28) versus 43% (AV14); difference of 6.4% [90%CI:-6.1% to 19.0%], not meeting non-inferiority criteria. Patients with NPM1/ IDH2 mutations had higher CR rates with AV28 (60.9% vs. 33.3%), while CR rates were equivalent (45%) for other subgroups. Composite CR rates were 80.7% (AV28) versus 68.6% (AV14) and MRD negativity was similar (77.6% vs. 76.5%). Although AV28 had more frequent treatment interruptions, count recovery after C2, grade ≥3 adverse events and early mortality were similar. In conclusion, the study did not demonstrate non-inferiority of AV14 compared with AV28 during C1-2 in an unselected ND-AML cohort. However, as the confidence interval for the difference covered 0, CR rate for AV28 was not significantly different than AV14. Certain subgroups may benefit from prolonged Ven exposure, but these findings require validation in larger studies, especially as triplet regimens evolve.

Securing the modern power grid with hybrid deep learning against cyber threats in renewable-integrated smart grids

Strong broadband nonreciprocal thermal radiation in multilayer gradient magneto-optical structure

Nonreciprocal thermal radiation, which violates symmetric reciprocity of traditional Kirchhoff's law, possesses numerous potential applications such as infrared camouflage and thermophotovoltaics. Here, we propose a multilayer structure composed of periodically stacked gradient doping concentration InAs layers and dielectric layers. Numerical simulations indicate that the proposed structure achieves strong broadband nonreciprocal thermal radiation at 18–20 μm, with average absorption, emission, and nonreciprocal performance of 0.29, 0.92, and 0.63, respectively. Such outstanding performance is attributed to the coupling of the epsilon-near-zero effect and Fabry–Pérot resonance, which can be demonstrated in the magnetic field distribution. The effects of incident angle, structural parameters, and magnetic field on nonreciprocal performance are also analyzed. We believe that this work paves the way for broadband, nonreciprocal thermal radiation, opening up new opportunities in fields such as energy utilization.

Quantitative guiding of developmental cell fate patterns using a dynamical landscape model

During development, cells gradually assume specialized fates via changes of transcriptional dynamics in thousands of genes. Landscape modeling approaches, which abstract from the underlying gene regulatory networks and reason in a low-dimensional phenotypic space, have been remarkably successful in explaining terminal fate outcomes. The success of these models also prompts their application toward inferring dynamic perturbations of multicellular patterning that alter cell fate outcomes in predictable ways, a task that is otherwise highly challenging due to the complex dynamics of the underlying gene circuits. Here, we accomplish this task by combining a landscape model for Caenorhabditis elegans vulval fate patterning with temporally controlled perturbations of EGF and Notch signaling in vivo using temperature-sensitive mutant alleles. We find that nonintuitive fate outcomes that emerge in combinations of these alleles at static temperature conditions through pathway epistasis are correctly predicted by the model. We then show that short pulses of signaling in these genetic backgrounds, delivered via temperature shifts, can be used to guide both the fraction of induced precursor cells and the specific fates they adopt with quantitative precision. Analysis of the underlying cellular landscapes indicates that cell fate guidance via pulses of signaling effectively redesigns the decision structure into one that has no equivalent in normal development, namely, a conversion of the three-way cell fate decision topology into two sequential binary fate decisions. Our results highlight the predictive power of landscape models and illustrate a method to quantitatively guide cell fate acquisition in a developmental context.