Artificial intelligence is now embedded in real medical research and, in some cases, real patient outcomes — from a melanoma vaccine trial to a fertility lab at Columbia University that used AI to find sperm a human eye had missed for two days.
On Aug. 19, Moderna and Merck announced positive topline results from a Phase 3 melanoma trial testing intismeran autogene, also called V940 or mRNA-4157, combined with Keytruda. The trial enrolled 1,137 people with high-risk melanoma that had already been surgically removed. The combination hit its primary goal of recurrence-free survival and a key secondary goal on distant metastasis-free survival. The companies say it’s the first positive Phase 3 result for both an individualized neoantigen therapy and an mRNA-based cancer treatment.
The process starts with a sample of the patient’s own tumor. Researchers analyze its mutations and use an algorithm to pick targets the immune system can be trained to recognize — an individualized therapy that can encode up to 34 neoantigens. Moderna says the V940 program uses integrated AI algorithms throughout development.
Only topline numbers are out so far. Full data are headed to an international medical meeting and to regulators, and the trial is still tracking overall survival. Earlier Phase 2b results, from a smaller group with longer follow-up, showed the combination cut recurrence or death risk by 49% and distant metastasis or death risk by 59% versus Keytruda alone. The FDA has not approved intismeran.
Old Drugs, New Uses
Dr. David Fajgenbaum co-founded the nonprofit Every Cure to hunt for treatments that already exist but aren’t being used where they could help. Every Cure’s 2025 annual report puts the number of recognized diseases worldwide at roughly 18,000, with only about 4,000 having an FDA-approved medication. The group uses AI to scan biomedical literature for overlooked drug-disease connections, generating tens of millions of predictions in under a day before researchers narrow down the most promising leads.
The federal Advanced Research Projects Agency for Health is funding a related project, MATRIX, which uses machine learning to flag FDA-approved drugs that might treat other diseases — leads that still require lab or clinical validation before anyone acts on them.
Fajgenbaum has seen the payoff of this kind of drug repurposing directly. Kaila Mabus developed multicentric Castleman disease at age 13 and remained severely ill despite chemotherapy. In 2020, doctors tried ruxolitinib — a drug approved for certain blood disorders, not Castleman disease. She began improving within months and was declared in remission in January 2021. AI wasn’t behind her specific treatment, but her case is the reason Every Cure wants to find those connections faster and at scale.
Finding a Single Cell in a Million Images
At Columbia University Fertility Center, researchers built the Sperm Tracking and Recovery system, or STAR, which pairs high-speed imaging with an AI detection model and microfluidics. It’s built for men with azoospermia or cryptozoospermia — conditions where sperm appear absent or exist in vanishingly small numbers. STAR processes about 1.1 million images an hour, with AI scanning each frame and a microfluidic mechanism isolating any sperm cell it confirms.
In one validation case, embryologists searched a sample by hand for two days and found nothing. STAR found 44 sperm in about an hour.
Columbia reports STAR achieved its first documented pregnancy in March 2025, for a couple who had tried to conceive for nearly two decades. It later resulted in a healthy delivery. But the technology isn’t a guarantee: Columbia says sperm are found in about 28% of patients with a prior azoospermia diagnosis, roughly 20% of mature eggs fertilize with STAR-recovered sperm, and about 18% of those fertilized eggs become good-quality embryos — lower rates than standard IVF or ICSI, reflecting how difficult these patients’ cases already were.
Predicting Heart Disease 15 Years Out
Researchers at the University of Hong Kong built CardiOmicScore, a deep-learning tool trained on large-scale data from the UK Biobank. It examines 2,920 circulating proteins, 168 metabolites and genomic information to estimate future risk of six cardiovascular conditions: coronary artery disease, stroke, heart failure, atrial fibrillation, peripheral artery disease and venous thromboembolism. The researchers found it improved risk prediction when layered on top of standard clinical data, and in some cases could flag elevated risk up to 15 years before symptoms showed up. It remains a research tool — not something available at a doctor’s office today.
Growing Tumors in a Dish to Test Drugs
At UCLA, scientists are growing tiny lab replicas of patient tumors called organoids, using a platform that combines 3D bioprinting, advanced imaging and AI. The AI processes the imaging data as organoids are exposed to different drugs, tracking thousands of them at once to see how different parts of a single tumor respond differently to treatment. The goal is matching therapies to an individual patient’s cancer, but the platform is still in development and validation.
Listening for Disease in the Voice
A Perspective published Sept. 4 in npj Digital Medicine looked at whether AI could detect voice biomarkers for ALS and Parkinson’s disease, since both cause measurable speech changes. The authors see potential for tracking ALS-related speech and swallowing changes specifically. But as of publication, no speech or voice-derived endpoint for either disease had been qualified by the FDA or the European Medicines Agency. One ALS speech-analytics platform holds FDA Breakthrough Device designation, which speeds up review — it is not the same as marketing authorization.
What Patients Should Ask
None of this means every AI-labeled tool in a doctor’s office deserves blind trust. A university research project and an FDA-cleared device are not the same thing, and patients should press for specifics before assuming otherwise.
- What does the AI actually do — analyze data for a doctor, or flag something for review?
- Does a doctor, specialist or lab professional check the AI’s findings before any decision is made?
- Has the FDA cleared or approved it, where that applies, and what research backs it up?
- How is your health data stored, who can access it, and could it be used to train the system?
The science here is real and moving fast. The melanoma trial results are encouraging but incomplete. Several of the tools described above are still experimental or limited to research settings. Patients deciding whether to trust AI with their care should demand the same evidence standard they’d expect from any other medical claim.
