How AI could transform cancer treatment and medicine
Researchers are increasingly turning to artificial intelligence to personalize cancer therapy and to discover new uses for drugs that are already approved. The technology can examine millions of microscopic images in a fraction of the time it would take a human, and it can sift through massive biomedical datasets to flag connections that might otherwise be missed.
One of the most advanced applications appeared on Aug. 19, when Moderna and Merck announced positive topline results from a Phase 3 trial of an individualized neoantigen cancer vaccine, intesimran autogene (also called V940 or mRNA‑4157), combined with the checkpoint inhibitor Keytruda. The study enrolled 1,137 patients who had undergone complete surgical removal of high‑risk melanoma. The combination met its primary endpoint of recurrence‑free survival and also achieved a key secondary endpoint measuring distant metastasis‑free survival. The companies said this was the first Phase 3 readout for an individualized neoantigen therapy and the first for an mRNA‑based cancer vaccine. The vaccine is created by taking a sample of the patient’s tumor, analyzing its unique mutations, and using an algorithm to select up to 34 neoantigens that the mRNA treatment then encodes. The goal is to train the immune system to recognize the specific molecular features of that patient’s cancer. While the companies described the findings as encouraging, they noted that the data are still topline; the full results will be presented at a major medical conference and submitted to regulators, and the trial will continue to follow overall survival. The U.S. Food and Drug Administration has not yet approved intesimran for melanoma treatment.
Earlier, smaller Phase 2b work with longer follow‑up showed that the same combination reduced the risk of recurrence or death by 49 % and the risk of distant metastasis or death by 59 % compared with Keytruda alone. Those results came from a much smaller cohort, underscoring the importance of the larger Phase 3 study.
Beyond oncology, a nonprofit called Every Cure, founded by Dr. David Fajgenbaum, is using AI to match existing FDA‑approved medicines with diseases that lack effective therapies. The organization’s 2025 report notes that roughly 18,000 diseases are recognized worldwide, yet only about 4,000 have approved treatments. Every Cure’s platform scans biomedical literature and generates tens of millions of potential drug‑disease connections in less than a day, allowing researchers to prioritize the most promising candidates for laboratory or clinical validation. The initiative is supported by the federal agency ARPA‑H through a project named MATRIX, which employs machine learning to predict which approved drugs might treat other conditions. AI does not prove efficacy; it simply helps scientists decide where to look next.
A concrete illustration of the value of repurposing drugs emerged in the case of Kaila Mabus, who developed multicentric Castleman disease as a teenager. Despite chemotherapy, she remained severely ill until doctors tried ruxolitinib — a medication approved for certain blood disorders but not for Castleman disease. Within months she began to improve and was declared in remission in January 2021. While AI was not the tool that identified ruxolitinib, the case highlighted why faster discovery of such matches could be life‑changing.
In the realm of reproductive medicine, Columbia University Fertility Center has deployed an AI‑driven system called STAR (Sperm Tracking and Recovery). The technology combines high‑speed imaging with a deep‑learning detection model and microfluidics to locate sperm that may be absent or present in minute numbers in azoospermia or cryptozoospermia patients. STAR can process roughly 1.1 million images each hour, and its AI examines each frame for sperm cells. When a cell is identified, a microfluidic device isolates it for use in fertility treatment or cryopreservation. In a validation sample, embryologists had searched for two days without success; STAR located 44 sperm in about an hour. The first pregnancy reported from the system occurred in March 2025, resulting in a healthy delivery after nearly two decades of infertility attempts. Columbia reports that sperm were recovered in 28 % of patients previously diagnosed with azoospermia, that about 20 % of mature eggs fertilize with the recovered sperm, and that roughly 18 % of those fertilized eggs develop into high‑quality embryos for transfer or freezing — rates lower than those seen with conventional IVF or ICSI but notable given the difficulty of the cases.
Another AI‑enabled development comes from researchers at the University of Hong Kong, who created CardiOmicScore. The system analyzes a broad panel of blood‑based proteins, metabolites and genomic data using deep learning to estimate a patient’s future risk for six cardiovascular conditions, including coronary artery disease, stroke, heart failure, atrial fibrillation, peripheral artery disease and venous thromboembolism. When combined with standard clinical information, CardiOmicScore improves risk prediction and has been shown to flag elevated risk up to 15 years before symptoms appear. The tool remains in the research phase and is not yet available for routine clinical use.
At UCLA, scientists are building tiny, laboratory‑grown replicas of individual patients’ tumors — so‑called organoids — using 3D bioprinting, advanced imaging and AI‑driven image analysis. The platform can monitor thousands of organoids simultaneously, testing how different tumor regions respond to various drugs. By revealing how cancer cells within the same patient may react differently to therapy, the approach aims to guide more precise treatment selections. The system is still being validated and has not yet been adopted in clinical practice.
Early work on voice biomarkers also suggests that AI may help monitor neurodegenerative diseases. A perspective published in npj Digital Medicine examined speech patterns in patients with amyotrophic lateral sclerosis (ALS) and Parkinson’s disease, noting that AI could potentially detect subtle changes related to disease progression, especially in speech and swallowing functions. While one ALS speech‑analysis platform has received FDA Breakthrough Device designation — an status that can accelerate regulatory review — it has not yet obtained marketing authorization, and no speech‑derived endpoint for these conditions has been approved by the FDA or the European Medicines Agency.
Across all these areas, experts stress that patients should understand the role AI plays in their care. Questions that can help assess the technology include whether a qualified professional reviews the AI’s findings before any treatment decision, whether the system has received regulatory clearance, what type of evidence supports its use, and how personal health data are stored and protected. The phrase “AI‑powered” can describe a wide spectrum of capabilities, so seeking a clear explanation of the technology’s function is advisable.
The growing integration of AI into medicine promises faster discovery, more personalized options and the potential to repurpose existing drugs for new indications. Yet the technology remains at varying stages of development, and many applications are still experimental or limited to specialized settings. As the melanoma vaccine trial and other advances demonstrate, rigorous evaluation and transparent communication are essential to ensure that AI‑driven innovations truly translate into meaningful health benefits for patients.
