The story archive

Every discovery.
A little clearer.

Browse the collection, follow a topic, or find the story that sparked your curiosity.

Topic

500 published stories

Digital Health

Fast but flawed: How DeepSeek AI performs in breast cancer care

Artificial intelligence is moving fast, and doctors are testing whether popular models can handle complex medical decisions. In a recent study, researchers evaluated whether the AI model DeepSeek could design treatment plans for breast cancer patients. They gathered anonymized medical records from 163 patients, including details like tumor characteristics and past therapies, and fed them into the AI using its "Deep Thinking R1" mode. The goal was to see if the machine could match the expertise of a human multidisciplinary team, which is a group of specialists who collaborate to guide patient care. The AI proved to be remarkably fast, generating a personalized treatment plan for each patient in under 20 seconds. However, the accuracy left a lot to be desired. Two senior clinical experts reviewed the results and found that only 25.2 percent of the AI-generated plans perfectly matched the regimens created by the human team. When researchers looked at a specific group of 96 patients to evaluate the AI as a decision support tool, they found an error rate of nearly 47 percent. The authors caution that this is a small, early-stage study conducted at a single hospital over just a two-week period. Still, the findings clearly show that the AI makes significantly different choices than human experts and requires strict professional oversight.

PubMed · 2 min read

Oncology

New Tool Predicts Brain Metastasis Risk in Breast Cancer

Breast cancer can sometimes spread to the brain, a severe complication known as brain metastasis. To help doctors anticipate this risk, researchers have developed a new predictive tool called a nomogram. This model functions like a customized calculator, combining a patient's unique biological markers, medical images, and clinical history to estimate their likelihood of developing brain tumors. The research team analyzed patient data to identify the most critical warning signs. They found that several specific factors independently increase the risk of brain metastasis. These include the number of tumors, the overall clinical stage of the cancer, and whether cancer cells have invaded the nearby blood or lymph vessels, a condition called lymphovascular invasion. The tool also looks at the Ki-67 index, which is a measure of how fast the cancer cells are growing and dividing, alongside the microscopic appearance and grade of the tumor cells. To ensure the tool works reliably, the scientists tested it on both the original group of 346 patients from a single hospital and a much larger external database of nearly 1,500 patients. Across all tests, the model exhibited robust predictive accuracy, reliably distinguishing between patients who would and would not experience brain metastasis. However, researchers note that because the initial training data came from a relatively small group at one institution, the tool will need to be tested across multiple clinical centers before it becomes a standard part of patient care.

PubMed · 2 min read

Oncology

How PET Scans Change Treatment for One in Five Abdominal Cancers

When doctors treat cancers of the abdomen, they rely on a group of experts known as a multidisciplinary team to build a comprehensive treatment plan. To get the best possible picture of the disease, these teams often use a highly detailed imaging tool called an 18-FDG PET/CT scan. This technology helps doctors spot active cancer cells that might be hiding from standard tests. A recent review of multiple clinical studies set out to discover just how often these specific scans actually change a patient's medical journey. By analyzing data from previous research, scientists found that these advanced scans altered the treatment plans for about 21.8 percent of adult patients with abdominal cancers. That means roughly one in five people had their medical strategy shifted directly because of the scan results. The changes made by the medical teams were often significant. Because the scans frequently revealed the hidden spread of the disease, doctors used the new information to add or cancel planned surgeries. In other cases, they modified how extensive a surgery needed to be. The scans also prompted doctors to shift the entire goal of treatment, moving between strategies meant to cure the disease and palliative care designed to manage symptoms and improve comfort.

PubMed · 2 min read

Oncology

Robotic Surgery Offers Safer Rectal Cancer Treatment for Younger Patients

When younger adults are diagnosed with rectal cancer, surgeons often rely on minimally invasive techniques to remove the tumor. For a common procedure called a low anterior resection, doctors typically choose between two methods: laparoscopy, which uses rigid tools and a miniature camera, and robotic surgery, where the surgeon controls highly flexible robotic arms from a console. A recent review evaluated how patients under the age of 65 fare with these two different surgical approaches. The findings suggest that robotic surgery offers several major advantages over traditional laparoscopy. Although the robotic procedures took significantly longer to complete, patients experienced fewer overall complications. They were also less likely to have their operation unexpectedly converted into a traditional open surgery. Crucially, the robotic approach was linked to lower death rates within the first thirty days after surgery and allowed patients to leave the hospital sooner. When it comes to directly treating the cancer, the robotic method showed a distinct technical benefit. Surgeons using the robot had higher success rates in performing a complete Total Mesorectal Excision, a critical technique that involves removing the rectum and its surrounding tissue to keep the cancer from spreading. Even with this technical edge, the long-term cancer outcomes, including overall survival and the chances of the cancer returning, were comparable between both surgical options.

PubMed · 2 min read

Digital Health

A popular AI chatbot for doctors is setting its sights on hospitals

Over the past few years, a health technology company called OpenEvidence has captured the attention of the medical community by offering a free artificial intelligence chatbot. The tool is powered by a large language model, which is a type of artificial intelligence trained on vast amounts of text to understand and generate human-like responses. OpenEvidence designed its specific model to help doctors navigate complex patient care decisions. Rather than trying to sell this software to hospital administrators, the company offered it directly to medical professionals. The approach was a massive hit. OpenEvidence claims that about 650,000 active physicians in the United States now use the platform, with medical trainees being particularly frequent users. Now, the company is attempting a major strategic shift. After building a huge user base of individual doctors, OpenEvidence is pivoting to target large hospitals and health institutions directly.

STAT News · 1 min read

Gene Therapy

New RNA upgrades make prime gene editing highly efficient

Prime editing is a highly precise genetic tool that allows scientists to rewrite DNA without making dangerous double-strand breaks in the genetic code. To do this, the system relies on a guide molecule made of RNA, which acts like a biological GPS to find the exact spot in the genome that needs fixing. However, these RNA guides are fragile and can break down inside cells before they finish the job. To protect them, scientists attach tiny stabilizing structures, known as motifs, to the end of the RNA guide. To find the best possible stabilizers, researchers developed a new high-throughput testing system called PE-PRISM. This tool allowed them to rapidly screen 2,858 different RNA motifs inside human cells. Through multiple rounds of testing and optimization, the research team identified three engineered winners named tevo2.0, eHAV, and eSBRMV1-A. When put to the test, these newly upgraded motifs completely outperformed the current standard stabilizer. In experiments designed to correct hundreds of disease-associated genetic mutations, the top new motifs proved more efficient than the older version over 90 percent of the time. The researchers successfully used these stabilized guides to correct mutations in human cells as well as inside the brains and livers of living mice.

Nature Biotechnology · 2 min read

Genomics

How an Alzheimer's Gene and Arsenic Team Up to Harm Brain Cells

Scientists have long known that Alzheimer's disease is driven by a mix of genetics and environmental factors, but they are just beginning to understand exactly how these forces interact to damage the human brain. A new study reveals that a specific genetic variation known as APOE4, which is famously linked to a higher risk of late-onset Alzheimer's, makes certain brain cells uniquely vulnerable to arsenic, a common environmental toxin. To explore this connection, researchers used a gene-editing tool called CRISPR to create lab-grown human microglia. Microglia act as the dedicated cleanup crew of the brain, constantly sweeping away cellular waste and defending against damage. By engineering these cells to carry either the high-risk APOE4 gene or the standard APOE3 gene, the team could compare how each group reacted to arsenite, a chemical form of arsenic. The results were striking. The cleanup cells carrying the APOE4 gene were much less likely to survive arsenic exposure. While the concentration of the toxin itself caused the biggest changes in how the cells behaved, the APOE4 mutation played a critical secondary role, specifically weakening the cells' ability to handle oxidative stress and clear away debris. The study also highlighted the impact on mitochondria, the tiny energy-producing powerhouses inside cells. Even before being exposed to toxins, the APOE4 brain cells had dysfunctional mitochondria. When introduced to low doses of arsenic, this preexisting damage grew significantly worse, causing the mitochondria to swell and lose their electrical charge. However, scientists note an important caveat. This research was conducted using early-stage, lab-grown cells rather than living human patients. While it does not yet provide clinical evidence, it offers a crucial new framework for studying how toxic environments might accelerate Alzheimer's disease in genetically susceptible people.

bioRxiv · 2 min read

Neurotech

Cracking the hidden code of brain cell bursts

Brain cells communicate by firing off tiny electrical signals called action potentials, or "spikes." For a long time, scientists have tried to understand this brain chatter by counting how many spikes happen over a given period, a method known as analyzing the rate code. However, relying solely on the overall firing rate misses crucial information hidden in the precise timing of these signals, especially when cells fire in rapid, clustered bursts. To solve this blind spot, researchers have developed a new mathematical tool designed to dig deeper into the data. Using a technique called factor analysis, this new method successfully untangles the hidden patterns driving these signals. Crucially, it separates the information encoded in rapid bursts from the information carried by single, isolated spikes. The researchers demonstrated that looking only at the standard firing rate completely obscures the true patterns underlying these neural bursts. The team tested their new framework on both computer simulations and experimental data. The tool accurately figured out the hidden structures of the data and identified when burst coding was taking place, all without needing outside clues.

bioRxiv · 1 min read

Neurotech

Scientists discover how the chemical GABA regulates brain timing

Our brains are constantly processing information at different speeds. Early sensory areas react quickly to the world around us, while other regions hold onto information longer to help us with complex tasks like working memory. At the same time, the brain organizes itself into large networks that rapidly switch on and off. Scientists have long suspected that this intricate sense of timing is controlled by a delicate balance between different chemical messengers. To find out exactly how this works, researchers focused on two main types of chemical receptors: GABA, which generally acts as a brake to inhibit brain activity, and NMDA, which excites nerve cells. They gave medications to alter the activity of these receptors in sixty healthy male participants. The researchers then mapped the brain activity of the volunteers at rest using magnetoencephalography, a noninvasive scanner that measures the magnetic fields produced by electrical signals in the brain. The results were clear. Boosting GABA activity directly prolonged the processing timescales of brain cells across multiple regions. This effect was most obvious in two specific areas. It increased the probability of the brain entering the frontal default mode network, which is typically linked to inward-focused thought, while reducing the occurrence of the dorsal attention network, which helps us focus on the outside world. Altering NMDA receptors, by contrast, produced no significant changes to the temporal organization of the brain.

bioRxiv · 2 min read

Genomics

Scientists Discover Hybrid Brain Cells Linked to Multiple Sclerosis

The human brain relies on a highly specialized workforce to stay healthy. For a long time, scientists neatly categorized these cellular workers. Microglia act as the immune and cleanup crew, while oligodendrocytes manufacture myelin, the protective insulation that coats nerve fibers. Now, researchers have discovered a fascinating hybrid. By looking at the genetic instructions of individual cells, they identified dual-phenotype microglia, or DPM. These unique cells simultaneously show the traits of both the immune cleanup crew and the insulation manufacturers. The research team found that these hybrid cells come in two distinct varieties, each with its own job and location. The first type, called myelin-associated DPM, is scattered throughout the entire brain. Controlled by a genetic switch known as Sox10, these highly branched cells help maintain nerve insulation. The second type, neuron-associated DPM, operates quite differently. Governed by a different genetic controller called Glis2, these cells cluster tightly in the hippocampus, a brain region crucial for memory. There, they help regulate the delicate communication networks between neurons. To make this discovery, scientists used a technique called single-cell transcriptomics, which allows them to read the exact genetic activity of individual cells alongside advanced physical mapping tools. While the findings are exciting, the researchers note that this is still early-stage preclinical work. The current study clearly maps out the genetics and physical locations of these hybrid cells, but further research in living models is required to fully confirm how they operate in a functioning brain.

bioRxiv · 2 min read

Neurotech

How the Brain Balances Biological Noise to Reliably See the World

When we look at the world, our brain cells fire off electrical signals called spikes to process what we see. However, these signals are naturally noisy. They vary from moment to moment, even when we are looking at the exact same object. A lingering mystery in biology is how the brain builds a reliable, consistent picture of the sensory world despite this constant, trial-to-trial fluctuation. Researchers studying the primary visual cortex of mice discovered a surprisingly simple mathematical rule behind this noise. They found that the average number of spikes and the variability of those spikes are perfectly linked in a specific ratio, displaying a slope of one on a logarithmic scale. To understand why this matters, the scientists used computer models to tweak this relationship. They found that the naturally occurring ratio hits a sweet spot. It minimizes the overlap between different visual inputs while keeping the sensory representations continuous. When the team simulated brain networks, they noticed that larger clusters of locally balanced neurons naturally produce this highly robust processing environment. If the brain instead used a system with uniform variability across all neurons, a slope of zero, it would process information more efficiently but would be much more fragile. Essentially, the brain sacrifices a bit of processing efficiency to ensure our visual perception remains strong and reliable.

bioRxiv · 2 min read

Drug Development

New Cell Model Helps Discover Drugs for Chemo Nerve Damage

Chemotherapy is a crucial treatment for cancer, but it frequently causes severe side effects. One of the most difficult is chemotherapy-induced peripheral neuropathy, a condition that damages the nerves and causes debilitating pain and numbness. Because scientists have lacked good laboratory models of human nerves, finding treatments to prevent this damage has been challenging. To solve this, researchers built a new testing platform using human induced pluripotent stem cells. These are regular adult cells that have been reprogrammed to grow into human sensory nerves. This gave the team a scalable and reliable way to study how paclitaxel, a common chemotherapy drug, causes nerve toxicity and damage. With this new cellular testing ground, the research team went hunting for protective drugs. They rapidly screened a library of 192 compounds known as kinase inhibitors, which are drugs that block specific proteins involved in cell communication. Out of this batch, they identified 19 compounds that successfully shielded the nerves from chemotherapy damage. By genetically altering the cells, they discovered that protecting the nerves specifically required blocking a combination of three enzymes at the same time: MAP4K4, MINK1, and TNIK. To confirm their findings, the researchers used targeted drugs to block those three specific enzymes. This approach successfully stopped nerve degeneration in their human cell models. Furthermore, the targeted treatment preserved nerve health and fiber density in a living mouse model experiencing chemotherapy-induced nerve damage.

PubMed · 2 min read