Volume 393 | Issue 6806 | 2 July 2026 — Summary of Science publications, focusing on content relevant to AI and life sciences.

🌳 COVER STORY — Height Doesn't Impair the Hydraulic System of the Tallest Tropical Dipterocarp Trees

Research Article · pp. 60–64 · DOI: 10.1126/science.aea9013

The tallest 1% of trees store more than half the carbon in the world's forests, and conventional theory holds that height limits water transport, making giants more vulnerable to drought. Studying Southeast Asian dipterocarps — the tallest tropical tree family, including Koompassia excelsa — across the 2023–2024 El Niño drought in Borneo, the authors found that these trees compensate hydraulically as they grow, preserving water transport despite their extreme height. The finding suggests greater drought resilience in these carbon-dense giants than previously assumed.

Science

🧬 Deep Learning-Designed "Protein Skirts" Solubilize Membrane Proteins — Mihaljević, Kim, Baker et al.

Research Article · DOI: 10.1126/science.adr3817

From David Baker's Institute for Protein Design: membrane proteins are central to signaling, infection, and drug response, but their hydrophobic surfaces make them notoriously hard to produce and study outside a lipid bilayer. The team used a deep learning-based design pipeline to generate target-specific, amphipathic "protein wraps" that solubilize transmembrane proteins by encasing their membrane-spanning regions. The resulting complexes fold correctly and are compatible with structural determination and biochemical assays — a generalizable tool with direct implications for antibody, vaccine, and structure-based drug design against integral membrane targets.

Science · bioRxiv preprint

🤖 Instead of Banning AI, I Made a Classroom Contract With My Students — Working Life Essay

Feature · Vol. 393, Issue 6806

A computer scientist-turned-educator recounts confronting a student who passed off AI-generated work as his own, and argues that an outright ban misses the point. Drawing on two decades of AI experience — including building fraud-detection and marketing-optimization systems — the author describes negotiating explicit, individualized agreements with students about acceptable AI use rather than issuing blanket prohibitions. A timely, first-person contribution to the ongoing debate over generative AI in scientific and technical education.

Science

🫀 Zonated Mechanosensing by PIEZO1 Controls Liver Regeneration

Research Article · DOI: 10.1126/science.aef0825

The liver's regenerative capacity depends on spatially organized proliferation, but how mechanical cues coordinate this zonation was unclear. This study shows that the mechanosensitive ion channel PIEZO1 senses tissue tension in a zone-specific manner in mice, directly linking physical forces to the liver's decision of when and where to regenerate after injury — a mechanobiology finding with potential relevance to regenerative medicine and liver disease modeling.

Science

🍽️ A Dietary Switch Promotes Sensory Neuron–Dependent Cancer-Associated Cachexia — Cross et al.

Research Article · pp. 90–97 · DOI: 10.1126/science.adz4196

In a subset of Lkb1-mutant lung cancer models prone to cachexia, a high-calorie, high-fat diet paradoxically triggered reduced appetite and weight loss. The tumor-derived lipid signal prostaglandin E2 (PGE2) was found to act locally in the lung to drive this response via sensory nerve signaling to the brain. Blocking PGE2 production or silencing the relevant sensory nerves reduced cachexia in mice, pointing to the peripheral nervous system as a potential therapeutic target (see accompanying Perspective by Gültekin & Vander Heiden, pp. 33–34).

Science

🧠 A Sub-10-Millisecond Neural Dynamical System Based on Phase-Change Memristors

Research Article · pp. 105–112 · DOI: 10.1126/science.aee6277

Phase-change memristors (PCMs) are attractive for analog, in-memory AI computation but suffer from conductance drift caused by structural relaxation. This work presents a neural dynamical system built on PCM hardware capable of adaptive, differentiable modeling of continuous deformation fields at sub-10-millisecond latency — a hardware-level advance relevant to real-time physical-world modeling and edge AI applications, including potential downstream use in fast biological simulation and sensing.

Science

Issue theme in brief: The clearest AI/life-sciences crossover this issue is the Baker lab's deep learning-designed membrane protein solubilizers — a direct extension of the de novo protein design toolkit (RFdiffusion/ProteinMPNN lineage) into a long-standing structural biology bottleneck. The classroom-AI essay and phase-change memristor hardware paper round out the AI angle, while the dipterocarp cover story, PIEZO1 liver regeneration, and cachexia papers are the standout life-sciences research articles.