Researchers turn to AI to decipher a devastating neurological condition

For individuals with amyotrophic lateral sclerosis (ALS), the diagnostic journey is typically a harrowing and expensive year-plus ordeal involving at least three healthcare professionals.

ALS, a fatal condition with few options for treatment, slowly robs patients of their ability to move and communicate. Diagnosing the condition can be extremely difficult, according to the ALS Association, a nonprofit group that supports ALS research.

Symptoms present differently in different patients, and rates of progression can vary. The standard questionnaire for measuring that progression is subjective by nature.

“If you’ve met one person with ALS, you’ve met one person with ALS,” Kuldip Dave, SVP of research at the ALS Association, told Morning Brew. “ALS looks different in every patient.”

Researchers are turning to tech to decipher the disease’s underlying causes. The hope is tools like AI, robots, and digital twins might help them identify biomarkers, or molecular fingerprints, that are specific to ALS. In theory, that would enable faster diagnoses and the development of much-needed medications, among other benefits.

“We don’t fully understand the biology as to why [ALS] happens. We don’t fully understand the biomarkers that would tell us whether somebody has ALS or diagnostic or prognostic biomarkers. Drug development is really difficult in ALS because of the variability,” Dave said. “If we can use AI to help us make it less complex, I think that’s the promise I see for AI.”

Last year, for example, a study published in the journal Nature Medicine identified dozens of proteins that were more abundant in the blood of patients with ALS, compared to healthy controls. The researchers then used those protein signatures to develop a machine learning model to diagnose ALS before symptoms appeared in their cohort of patients. The results were encouraging, according to Dave, but it’s a small study and the model will need to be validated further.

Steven Finkbeiner, senior investigator and director of the Center for Systems and Therapeutics at the University of California, San Francisco-affiliated Gladstone Institutes, is trying to speed up the discovery of the molecular missteps that lead to ALS by using thousands of individual neurons and what he calls a “thinking microscope.”

The patented robotic system was inspired, in part, by AlphaGo, the AI system developed by Google DeepMind to master the board game Go. As with AlphaGo, Finkbeiner wants to eventually use reinforcement learning—a type of AI capable of learning from experience—to power the smart microscope. The microscope currently uses a computer vision system and a large language model (LLM) to analyze the many images the microscope takes, though Finkbeiner doesn’t think LLMs are the most efficient setup in the long run, as some of the detail gets lost in translation from image to text, he said.

To do these “high-throughput” experiments, Finkbeiner’s team uses 384-well plates, rectangular containers that resemble miniature cupcake molds. Each well contains a “soup” of neurons. Finkbeiner says his robotic microscope can target individual cells to tweak the configuration of proteins known to be involved in disease progression. Some cells have been genetically altered to mimic certain features of ALS or other neurodegenerative disorders. Others come directly from patients with these disorders. (His lab also studies dementia, Huntington’s, Parkinson’s, and Alzheimer’s.)

One recent preliminary experiment, for example, involved modifying the amount of oxidative stress that neurons were exposed to. His team found that as oxidative stress increased, so did cell death. That was not surprising; oxidative stress is a common theme across neurodegenerative disorders, including ALS. But cells that experienced low levels of oxidative stress in the experiment lived longer than those exposed to none. What the optimal levels of stress are remains a mystery. These data have not yet been peer-reviewed.

Using the data from experiments on thousands of neurons, “my hope is that we can really generate a very refined frontier model of these different diseases that’s queryable and can give us more reliable direction scientifically,” Finkbeiner said. “So I’m hopeful that it will both accelerate the work and the product will be more reliable.”

Tracking deterioration

EverythingALS, a patient-advocacy nonprofit focused on leveraging tech to improve ALS care, is building a tool to evaluate how patients’ ability to speak declines as the disease progresses. The system leverages thousands of voice samples and speech pathologist evaluations to train AI models based on attention—the same context-understanding mechanism that powers LLMs.

An early version of the experimental model was described in a May 2025 paper published in the journal Digital Medicine. The authors claim the model offers a greater degree of interpretability as compared to other deep learning approaches.

EverythingALS also recently rolled out a tool that uses AI to match ALS patients to clinical trials they might qualify for. The nonprofit’s tool, called Sava AI, is meant to make the experience conversational and bring in doctors to annotate descriptions of the trials.

“Making the information accessible at the right time for the patients is so critical,” Indu Navar, CEO and founder of the nonprofit EverythingALS, told us. “Otherwise, the one chance you have to maybe slow it down or recover is taken away.”

Accessibility is important because patients need to be within a certain window of time from the onset of their symptoms in order to qualify for trials.

“There is an inclusion criteria: You need to be 16 months to 18 months from the [onset] of your symptoms to enroll, and it takes about 18 months to 24 months to get diagnosed,” Navar said. “So a lot of people are going to be outside the window if they don’t get to know what trials they’re eligible for immediately.”

Some researchers have also begun to use AI-powered digital twins to simulate placebo groups in certain trials, according to Dave.

This sort of digital twin placebo group is one of four major areas where Dave sees AI potentially having a major impact on ALS research. The other three are using AI on large omics datasets, determining biomarkers, and improving assistive technology, like voice banking that allows those living with the disease to preserve their speaking voice digitally.

“In a lot of these areas, they’re early. They’re small studies,” Dave said. “We need to see whether these will pan out. But if they do, they will change the way drug discovery happens, not just for ALS, but across the board for any disease.”