Scientists use AI and optogenetics to make major breakthrough in Parkinson’s treatment

A South Korean research group has reported in a new paper that artificial intelligence and light-based genetic tools can work together to diagnose and stop Parkinson's-like disease in mice. The work is a long way from being able to affect humans, but it gives some insight into how technology could revolutionize the diagnosis and treatment of neurodegenerative illnesses.

Parkinson's disease is a progressive brain disease that disrupts movement. Tremors, stiff muscles, slowness of movement, and difficulty with balance are common. The condition is caused by the loss of dopamine-producing cells in a small region of the brain known as the substantia nigra.

Despite decades of study devoted to it, no cure exists. What treatments are available today will palliate symptoms but not ward off the loss of neurons. For scientists, one of the biggest hurdles has been how to measure small behavioral changes in laboratory animals. Easy tests like walking along beams or rotating rods have a tendency to overlook early warning signs.

In the new research, a team at KAIST and the Institute for Basic Science created a mouse model of Parkinson's with a human protein named alpha-synuclein. A mutation of the protein, A53T, occurs with inherited Parkinson's in humans. The protein clumps up and kills off neurons when overexpressed in the brain, and this leads to Parkinson's-like symptoms.

Mice were grouped and injected with control viruses or varying amounts of A53T protein in the substantia nigra. In ten weeks, the group that received the higher dose severely suffered from movement issues. Their dopamine-making neurons plummeted to roughly one-fifth normal levels, and striatal fibers—nerve tracts important for motor function—dropped below a third.

The actual breakthrough came when scientists used artificial intelligence to investigate the movement of mice. Prying open windows in a dimly lit room with several cameras, they constructed 3D skeletons of freely roaming mice using software called AVATAR. From these, they quantified hundreds of features—everything from the tilt of a paw to a beat in a chest movement.

Machine learning models were learned on these data and an Extreme Gradient Boosting model performed best. The team created a new measure called the AI-predicted Parkinson's score, or APS. The APS caught the number of movement clips that were labeled as Parkinson's-like.

By week two, the APS was already able to differentiate diseased from healthy mice—long before standard tests had picked up on clear differences. At ten weeks, severely affected mice had scores on the APS of over 85 percent, while controls were still at about 12 percent. The APS also picked up on neuron loss better than any standard test, and therefore it was a more state-of-the-art tool for diagnosis.

Because the system was clear, researchers could pinpoint which characteristics were most significant. Coordination of limbs was high on the list. Diseased mice exhibited tighter hand spacing, wider bases, and more asymmetrical foot placement. Their hind-leg-standing rearing movements were slower and less vigorous. Posture also altered—stiffened bodies, slouched postures, and reduced speed of neck movement became significant markers.

These small signs, more than naked-eye detection, were the entry point to early and accurate identification.

Diagnosis was only half the puzzle. The team also explored treatment with an optogenetic device called optoRET. This light-controlled gene-based system offers light control over a receptor called c-RET, which favors survival of neurons. Instead of using drugs that need to travel all the way to the brain, optoRET activates protective pathways with light on demand.

Mutant A53T mice were treated with optoRET and subsequently subjected to blue light exposure at varied schedules—every day, every two weeks, or every other day. The findings varied as a function of disease severity.

In the mild group, light applied every other day or twice a week slowed disease development. None of these mice attained severe levels by ten weeks. They also regained some motor function: they walked beams more steadily, had up to 90 percent of their dopamine neurons remaining, and had nerve fiber density close to normal.

Daily light treatment was unexpectedly less effective, suggesting that periodic stimulation could be more effective than continuous stimulation.

AI analysis demonstrated how optoRET worked. In treated mice, only 16 percent of movements were still Parkinson's-like. Key features—specifically limb coordination—returned to healthier patterns. Their gait also improved, with fewer dragging steps and more stable turns. Even such trivial behaviors as chest movement entropy, a measure of variability, improved.

Simply put, light therapy restored both function and neuron survival when given at the appropriate rhythm.

To verify whether the AI system was specifically identifying Parkinson's and not just identifying generalized motor deterioration, researchers also examined mice with amyotrophic lateral sclerosis, or ALS. While the ALS mice also exhibited motor problems, their APS values were not elevated. Their activity patterns were quite different, affirming that the AI was identifying Parkinson's-specific patterns.

KAIST Professor Won Do Heo underscored the significance: "This is the world's first preclinical platform that has been implemented to connect the early diagnosis, treatment evaluation, and mechanism validation of Parkinson's disease using a combination of AI-driven behavioral analysis and optogenetics."

The study, led by postdoctoral researcher Dr. Bobae Hyeon and published in the journal Nature Communications, presents a proof-of-concept for a new research paradigm.

The research remains problematic. The mouse model carries just one kind of Parkinson's, and translating optogenetic tools into humans will be difficult. The exact molecular mechanism of optoRET's protection is unknown, and light delivery in human patients would be problematic. But success in the mouse model provides opportunities to make something better happen.

By bridging AI with biological control, the research points towards a direction for future earlier diagnosis and more effective treatments. Instead of drugs for everyone, future medications may be tailored, trigger only when the time is right, and be monitored in real time.

This study sets the stage for technologies that could potentially detect Parkinson's years ahead of tests, maybe years ahead of when the symptoms even exist. For patients, it could mean treatment sooner and better disease management.

The optogenetic results also imply treatments at the optimal time specific to the individual, where treatment is only given when it needs to be, eliminating side effects and maximizing efficacy.

For researchers, AI offers a powerful tool to monitor fine behaviors that can predict the onset of other neurological illnesses, opening new frontiers for personalized medicine.

Like these kind of feel good stories? Get The Brighter Side of News' newsletter.