How and why a tumor metastasizes — when a seemingly similar tumor doesn’t — remains one of the biggest questions in cancer research.
New research from György Marko-Varga’s lab at Lund University with Istvan Nemeth at the Szeged Clinical hospital, and Peter Horvath’s team at HUN-REN Biological Research Centre suggests a new way to chip away at this question with the help of AI combined with spatial proteomics to reveal the single-cell dynamics of different populations of cancer cells within the same tumor. Their approach offers a promising step up over traditional molecular analyses that view the entire tissue sample at once and miss the functional differences in how different cell types behave.
In their case report of a young woman with metastasized melanoma, the researchers identified two spatially distinct tumor cell populations in the original tumor. One, PT1, was an aggressive melanoma subtype that likely drove metastasis into the lungs and eventually the brain.
Their analysis also surfaced metabolic and kinase programs associated with the progression and treatment resistance, which suggested that targeted therapies may not work well for the patient, with metabolic inhibitors potentially offering greater gains for the patient.
“These findings demonstrate the potential of patient-specific AI-imaging interfaced with proteomic tumor mapping to move beyond conventional pathology, reveal actionable vulnerabilities, and guide more precise combination treatment strategies for advanced melanoma,” Marko-Varga told DDN.
Finding the path to metastasis
The research team’s new pipeline rests on the combination of deep learning-guided histopathological analysis with proteomics and laser microdissection. Crucially, the new insights from this morphological and molecular data was overlaid onto the timeline of the patient’s disease to elucidate which cell populations drive metastasis.
“To be able to reconstruct and connect localized tumor-cell regions at single-cell resolution, integrating protein expression with pathology-guided annotation and spatial indexing, was a major achievement,” said Marko-Varga. “This allowed us to explain tumor aggressiveness at a functional level and directly correlate morphological features with underlying molecular signaling mechanisms.”
These findings demonstrate the potential of patient-specific AI–proteomic tumor mapping to move beyond conventional pathology, reveal actionable vulnerabilities, and guide more precise combination treatment strategies for advanced melanoma.
— György Marko-Varga, Lund University
Specifically, the metabolic programming they identified all the way from original tumor to metastasis showed activation of both glycolysis and oxidative phosphorylation. At the same time, the metastatic tumors showed strong activation of MAPK (mitogen-activated protein kinase) signaling and PI3K–AKT–mTOR (phosphoinositide 3-kinase– protein kinase B–mechanistic target of rapamycin) pathways along with decreases in anti-tumor immunity — ultimately representing tumor progression into a largely immune-suppressed state.
“Collectively, these findings identify a convergence of metabolic rewiring, oncogenic pathway activation, and reduced anti-tumor immunity during melanoma progression,” noted Mark-Varga. “By resolving these mechanisms within spatially defined tumor-cell populations, we establish a direct link between single-cell morphology, tumor architecture, protein expression, and pathway activity, providing a framework for identifying actionable vulnerabilities driving melanoma metastasis.”
These findings also allowed the team to make more informed suggestions about which treatments might work best for this individual patient. “In particular, inhibitors targeting mTOR/PI3K signaling, EGFR, VEGFR, and microtubule dynamics in Patient Y case showed strong negative connectivity scores, consistent with our pathway-level findings indicating activation of AKT–mTOR and related proliferative signaling networks in the metastatic tumor,” explained Marko-Varga.
Paving the way for precision medicine
The new study offers a new roadmap for how to use AI to advance precision oncology — an area of continued interest for Big Pharma.
Current forecasts from the Oncology Precision Medicine Market Report 2026 suggest that the market will grow from $141.22 billion in 2026 to $226.55 billion by 2030. The recent positive results from the large-scale melanoma personalized vaccine trial from Moderna and Merck also reinforces this direction, and Marko-Varga said it represents a new opportunity for patients that will open up novel avenues for them to help select patients who have responded well in their future studies.
“Individualized cancer treatment is undoubtedly the way forward. As my team has studied many thousands of patient tumors over the years, it is clear that every tumor is biologically unique with individual architectures,” said Marko-Varga. He added that the bottom line is that “ultimately, each patient will likely require a dedicated and individualized combination of therapies, guided by the specific biological characteristics of their tumor.”
Marko-Varga’s team will soon publish similar results with an AI-based personalized medicine approach in two other patients.
“Our team’s ambition is to move beyond treating the tumor alone and toward a truly patient-centered precision oncology approach, in which science-driven evidence, clinical judgment, and the patient’s own experience collectively contribute to the treatment strategy,” he said.
