5 Science News Stories — October 22, 2026

AI Tests Cancer Drugs, Magnets Learn a New Trick, and Plastic Breaks Down in the Deep Sea

THURSDAY  •  OCTOBER 22, 2026

Five verified stories reported on October 21: AI-guided cancer-drug research, pregnancy immune-system digital twins, a new magnetic switching effect, NASA’s experimental 3D lidar, and enzyme-embedded plastics tested in the ocean.

Five October 21 research stories, from laboratory medicine and experimental memory physics to Earth mapping and marine materials. AI-generated editorial concept art, not documentary photographs.

01  /  ARTIFICIAL INTELLIGENCE + MEDICINE

An AI Researcher Finds a New Lead Against Drug-Resistant Prostate Cancer

UCLA Health announcement: October 21, 2026. Original paper: October 8, 2026.

Cancer rarely follows a simple script. A treatment can work for months or years, then the tumor finds another way to grow. That is the problem UCLA researchers tackled with MAESTRO, an AI-assisted system for finding drug combinations worth testing. Announced on October 21, the work is a promising demonstration of research acceleration, not a new treatment available to patients.

In advanced prostate cancer, a drug called enzalutamide blocks signals from the androgen receptor, a pathway many tumors use to survive. But cancer cells can activate alternative pathways and become resistant. Testing every possible partner drug would take enormous time and money. MAESTRO instead gives large language models carefully selected information about drugs and cancer biology, then asks them to rank possible combinations and explain their reasoning. Candidates are evaluated repeatedly, rather than trusted on the strength of a single AI answer.

The team screened 3,592 natural products computationally and chose 150 high-confidence candidates for laboratory testing. Of these, 23 showed signs of synergy with enzalutamide in three-dimensional clusters of prostate cancer cells. One combination, berberine plus enzalutamide, showed consistent synergy across four laboratory models with different resistance mechanisms. Follow-up measurements were compatible with changes in the AMPK and mTORC1 signaling pathways, which help regulate cellular energy and growth.

Why does this matter? The AI did not simply produce a list of fashionable molecules. It helped prioritize a manageable set of experiments and offered biological hypotheses that scientists could challenge. The researchers also supplied extra oxygen to the tumor-like cell clusters, reducing misleading cell death caused by the testing setup itself. That is an important reminder: better experiments matter as much as smarter algorithms.

The limits are substantial. These were cell-model results, not evidence that the combination helps people with cancer. Berberine is not a proven substitute or add-on for prescribed treatment, and drug interactions and safety still need proper assessment. Animal studies and patient-derived models are among the next steps. The broader story fits the shift toward AI-supported drug discovery, where a computer helps decide which hypotheses deserve scarce laboratory time.

PRIMARY SOURCE  Original study — Advanced Science   •   UCLA Health announcement

02  /  MEDICINE + DIGITAL BIOLOGY

A Digital Twin of Pregnancy Points to a Possible Way to Delay Early Labor

Stanford Medicine announcement and Science Advances publication: October 21, 2026.

A digital twin sounds like a virtual copy of a person. In medicine, it is more useful to think of it as a mathematical model of one part of the body: a system that can predict how biological signals may respond to an intervention. On October 21, Stanford Medicine researchers described SIMCO, a model of the immune changes that help regulate the timing of labor. It has highlighted a drug combination for further clinical testing.

Pregnancy involves a delicate immune balancing act. Early on, the mother's immune system adapts to support a developing fetus. Near the end, inflammatory signals help prepare the body for delivery. If this biological timetable advances too quickly, preterm labor can follow. Researchers trained SIMCO using responses measured in roughly 250 million immune cells and combined those data with information from 63 pregnancies to capture variation between individuals.

The team used these models to compare nine drug regimens. The strongest simulated result came from a combination of low-dose aspirin and lansoprazole, a medicine used to reduce stomach acid. In the model, the combination's predicted effect on delaying the immune progression toward labor was about 15 times that of aspirin alone. Researchers also observed delayed labor in an animal model of pregnancy-associated inflammation, and a human trial is being prepared.

That sounds dramatic, but the distinction between a modeled biological effect and a clinical benefit is crucial. 'Fifteen times' is not a measured 15-fold reduction in premature births among pregnant women. It describes an outcome in the researchers' computational system. No one should begin taking either drug during pregnancy on the basis of this report; medication decisions depend on a clinician's assessment of the specific pregnancy.

The larger possibility is a safer route to selecting which treatments deserve clinical trials. Instead of exposing large numbers of people to speculative therapies, scientists can use detailed cell measurements to narrow the list, test the underlying mechanisms and then ask whether the predictions survive real-world trials. This is an especially important test of AI in predictive medicine: a convincing simulation must ultimately improve outcomes for mothers and babies.

PRIMARY SOURCE  Stanford Medicine research announcement

03  /  PHYSICS + FUTURE COMPUTING

A New Magnetic Switching Effect Could Open a Different Road to Computer Memory

Northwestern Engineering announcement and Nature paper: October 21, 2026.

The computers we use every day depend on being able to store a bit and reliably change it from zero to one. Some advanced memory devices do that with magnetism. Now Northwestern-led researchers have demonstrated a previously unobserved way to switch a magnetic state using an electric current. Their October 21 Nature paper describes an effect called octupole-driven spin-transfer torque, or OTT.

Conventional magnetic memory often relies on ferromagnets, where many microscopic magnetic moments line up. Antiferromagnets are different: their moments tend to point in opposing directions, largely canceling the material's overall magnetization. That makes them interesting for dense and potentially fast electronics, but it also raises a puzzle. How can a current write information into a magnetic system whose net magnetization is essentially zero?

The researchers built tiny tunnel junctions with two layers of a material called PtMn3 separated by an extremely thin magnesium-oxide barrier. At room temperature, the devices showed electrical switching and a tunneling magnetoresistance signal of 363 percent. In plain language, the electrical resistance changed strongly enough between magnetic states that the researchers could distinguish them. The team developed a theoretical explanation involving a more complicated pattern of magnetic moments, described by an octupole rather than a simple magnetic dipole.

The result matters because useful memory needs both sides of the transaction: writing a state and reading it back. The experiment shows both can occur in this class of all-antiferromagnetic junction. It also gives physicists a new mechanism to investigate, rather than merely a clever way to package familiar magnetic components.

But a laboratory switching demonstration is not a ready-made memory chip. The researchers have not shown that these devices outperform commercial memory in energy use, endurance, cost or manufacturability. The reported 363 percent is a resistance contrast, not a speed improvement or energy saving. If future engineers can solve the practical constraints, the physics could support denser magnetic storage and potentially ultrafast spintronic components.

PRIMARY SOURCE  Original study — Nature   •   Northwestern Engineering announcement

04  /  SPACE TECHNOLOGY + EARTH OBSERVATION

NASA Wants to Turn Thin Laser Tracks Into Continuous 3D Maps of Earth

NASA technology update: October 21, 2026. Airborne prototype flight: November 2024.

Imagine trying to map a forest by drawing a few thin lines through it and guessing what lies between them. That is roughly the limitation of some current spaceborne laser mapping systems: they measure narrow ground tracks, leaving large gaps. In an October 21 update, NASA described CASALS, an experimental lidar instrument designed to collect broad, continuous three-dimensional swaths instead.

Lidar works by sending out laser pulses and timing their reflections. The returning light can reveal the height of trees, the shape of the ground and the layers of vegetation between them. CASALS changes how the laser is aimed. Rather than relying on moving mirrors, it rapidly shifts among 256 wavelengths around 1,040 nanometers. A tiny optical grating bends each wavelength in a slightly different direction, sweeping the pulses across the terrain below.

NASA also reports a detector quantum efficiency of about 80 percent, meaning the sensor converts a large fraction of returning photons into usable signals. That can help with darker surfaces and dense vegetation. An airborne prototype flew on a NASA King Air aircraft in November 2024. From roughly five kilometers up, it produced data with approximately 22-centimeter horizontal resolution, fine enough to distinguish branches and subtle changes in the ground.

Why should anyone care about more complete 3D maps? Forest structure influences estimates of carbon stored in vegetation. Elevation maps help researchers understand flood risk, erosion, changing landscapes and ecosystems. A system that measures the spaces between today's narrow laser tracks could provide a much richer picture of the planet, particularly where ground surveys are difficult.

The critical limitation is that CASALS has not yet been demonstrated in orbit. The 22-centimeter result belongs to the airborne prototype, not a promised satellite specification. Engineers still need to widen the scanning swath, add onboard processing and adapt the hardware for space. NASA plans further airborne work, including a campaign targeted for 2027. For now, the achievement is a promising flight-tested instrument, not a finished global mapping service.

PRIMARY SOURCE  NASA — CASALS instrument technology update

05  /  ENVIRONMENT + MATERIALS SCIENCE

Plastic With Enzymes Built Inside It Can Break Down on the Deep Seafloor

Original peer-reviewed Nature study published October 21, 2026.

A plastic bag marked 'compostable' does not necessarily disappear if it ends up in the ocean. Many compostable polymers need warmer, controlled conditions to break down efficiently, while deep water is cold and under enormous pressure. A Nature study published on October 21 explores a different strategy: putting the chemical tools for degradation inside the plastic from the beginning.

Researchers embedded small amounts of enzymes into several compostable polyesters, including PBS, PBSA, PCL and PBAT. Enzymes are proteins that speed up chemical reactions; here, they help cut long polymer chains into smaller molecules. The team formed thin films and tiny beads, then tested the materials in coastal waters and at deep-sea sites reaching 5,503 meters below the surface. The enzymes included a lipase and a cutinase chosen for their ability to act on these polymers.

The results were striking, but specific. Some enzyme-containing polyesters reached nearly complete degradation at a depth of 757 meters within five months. At the deepest site, degradation was measurable but slower. In coastal water, as little as 0.1 percent enzyme by weight accelerated breakdown for some materials while largely preserving their initial mechanical strength. Deep-sea experiments generally used higher enzyme loadings to cope with harsher conditions.

The attraction is obvious: a material could remain useful during its working life but become less persistent after entering water. That may be valuable for applications where accidental loss into the sea is difficult to eliminate. The researchers also examined tiny plastic-like beads to explore whether enzyme activity can continue after a material fragments.

Still, this is not a universal fix for plastic pollution. The work does not show that ordinary polyethylene bottles or conventional PET waste will suddenly dissolve. Degradation speed varies with polymer, depth, temperature and enzyme loading. Scientists also need to examine breakdown products, ecological effects, cost and real-world manufacturing at scale. Better materials could reduce future pollution, but preventing plastic from entering the ocean remains the first line of defense. Earlier Next Horizon coverage explored microplastics and their chemical behavior; this new paper addresses a different question: whether future materials can be engineered to break down sooner.

PRIMARY SOURCE  Original study — Nature