IntroductionLarge-batch training is widely used to scale multimodal neural networks that integrate heterogeneous inputs such as visual, textual, and physiological signals. However, increasing the batch size suppresses the stochastic fluctuations of mini-batch sampling, which can trap multimodal models in sharp, modality-dominant minima and produce a persistent generalization gap.MethodsTo address this problem, we propose Geometric Anisotropic Noise Injection (GANI), a curvature-aware optimization framework inspired by information geometry and multisensory integration. GANI decouples deterministic large-batch descent from stochastic geometric exploration. It approximates local curvature through an exponential moving average of first-order gradients and injects structured anisotropic noise during parameter updates, thereby restoring the geometry-aware exploration dynamics of small-batch stochastic gradient descent with linear computational complexity.ResultsTheoretical analysis shows that curvature-aligned stochasticity can accelerate escape from sharp modality-specific basins and guide parameters toward flatter regions. Experimental evaluations across multimodal benchmark settings and stress tests demonstrate that GANI reduces the generalization gap, improves convergence stability, bounds the maximum Hessian eigenvalue, and maintains stronger performance under visual noise and missing textual information than standard large-batch SGD and common adaptive optimizers.DiscussionBy linking optimization geometry with multimodal representation dynamics, GANI provides an efficient and interpretable mechanism for robust heterogeneous data processing. The framework offers potential value for uncertainty-aware multisensory integration, brain-inspired perception science, and scalable multimodal learning under noisy or incomplete sensory conditions.
SMILE: neural signal acquisition and intra-body transmission for facial nerve bypass—An acute feasibility study and proof-of-concept in a rat model
Facial paralysis is a disabling condition with severe functional and aesthetic consequences. Facial paralysis affects approximately 1.8% of individuals over their lifetime, with approximately 30% of affected patients developing persistent deficits; among these, patients with permanent flaccid paralysis and severe facial asymmetry do not resolve with pharmacological treatment and require surgical intervention. It is specifically this surgically relevant subgroup that represents the target population of the SMILE framework. The SMILE framework (bypaSs of a facial nerve lesion through intra-body biocoMpatIbLE communication technologies) validates the feasibility of the communication infrastructure required to establish a functional neural bypass link between the healthy side and a surgically reinnervated contralateral side of the face. This work presents a preliminary interdisciplinary experimental and engineering approach underlying the SMILE framework, combining neurophysiological validation in 15 adult Wistar rats with the design of ultra-low-power intra-body communication links based on galvanic coupling (GC) and ultrasound (US). Microsurgical cuff electrodes were implanted around the buccal branch of the healthy facial nerve to record odor-evoked motor outputs. ENG signals recorded from the intact buccal branch of the facial nerve on one side were transmitted to the contralateral side, across the animal’s facehead. Engineering evaluations demonstrated robust transmission capabilities, with the GC link achieving a mean equivalent SNR of 18.2 ± 0.6 dB and a mean normalized cross-correlation of r = 0.72 ± 0.09 between transmitted and reconstructed ENG signals, with MSE on the order of 10−2. The US link achieved an equivalent SNR around 12 dB with MSE around 5 · 10−2, supporting the feasibility of intra-body neural signal relay through biological tissues.
PANoptosis in neurological disorders: from inflammatory cell death mechanisms to neuroprotective strategies
PANoptosis is now regarded as an inflammatory form of programmed cell death (PCD). It reflects the coordinated involvement of apoptosis, pyroptosis, and necroptosis, usually through the PANoptosome in a shared pathological environment. This concept may be especially useful in neurological diseases. It helps explain why neuronal death, sustained inflammatory activation, and tissue injury often develop together and reinforce one another. Neural tissue is particularly sensitive to oxidative stress, mitochondrial dysfunction, immune-mediated inflammation, and blood-brain barrier disruption. These pathological changes are common in many forms of neural injury. Therefore, abnormal PANoptosis activation may provide a common mechanism linking different types of nervous system damage. This review summarizes the historical evolution, molecular mechanisms, disease-related roles, and intervention strategies of PANoptosis in neurological disorders. It focuses on PANoptosome assembly and key mechanistic nodes, including NOD-like receptor family pyrin domain-containing 3 (NLRP3), caspase-8, the receptor-interacting serine/threonine protein kinase 1 (RIPK1)/receptor-interacting serine/threonine protein kinase 3 (RIPK3)/mixed lineage kinase domain-like protein (MLKL) axis, gasdermin D (GSDMD), and Ninjurin 1 (NINJ1). It also highlights current translational limitations, such as disease heterogeneity, incomplete cell-specific validation, and insufficient clinical evidence.
Editor’s pick: Liberate Bio
Nature Biotechnology, Published online: 30 June 2026; doi:10.1038/s41587-026-03201-5
Each year, Nature Biotechnology highlights companies that received sizeable early-stage funding in the previous year. Liberate Bio is developing a lipid nanoparticle toolbox to deliver genetic medicine to previously inaccessible cells inside the body.
Editor’s pick: Trogenix
Nature Biotechnology, Published online: 30 June 2026; doi:10.1038/s41587-026-03200-6
Each year, Nature Biotechnology highlights companies that received sizeable early-stage funding in the previous year. Trogenix has developed a platform for generating ‘synthetic super-enhancers’ that could drive highly cell-specific gene therapy-based treatment of solid tumors such as glioblastoma.
Prime editing for precise genome engineering and modulation of fungal metabolism
Nature Biotechnology, Published online: 30 June 2026; doi:10.1038/s41587-026-03202-4
fPE7max efficiently installs base substitutions, insertions and deletions to modulate filamentous fungal metabolism.
Author Correction: Digital AVATAR therapy for distressing voices in psychosis: the phase 2/3 AVATAR2 trial
Nature Medicine, Published online: 30 June 2026; doi:10.1038/s41591-026-04540-1
Author Correction: Digital AVATAR therapy for distressing voices in psychosis: the phase 2/3 AVATAR2 trial
Multi-antigen-targeting T cells in pediatric central nervous system tumors: a phase 1 trial
Nature Medicine, Published online: 30 June 2026; doi:10.1038/s41591-026-04449-9
In the phase 1 ReMIND trial of tumor-associated antigen-specific T cells in patients with pediatric central nervous system tumors, treatment was generally well tolerated with one complete response and three long-term responders.
Clinical decision support in hematological malignancies using a case-grounded AI agent
Nature Medicine, Published online: 30 June 2026; doi:10.1038/s41591-026-04494-4
A locally deployable, case-grounded large language model agent achieved high concordance with hematology tumor board decisions across retrospective, external and prospective evaluations.
Connectomic evidence that ordered activity drives neuromuscular network formation
Nature Neuroscience, Published online: 30 June 2026; doi:10.1038/s41593-026-02344-7
Complete neuromuscular connectomes across development show that motor neuron recruitment order shapes both local and long-distance synapse elimination, leaving adult wiring as a structural record of the activity patterns that formed it.