A Korean research team has developed an artificial intelligence (AI) prediction model that can screen in advance, even before cancer treatment begins, for a deadly liver complication that can occur in pediatric patients receiving hematopoietic stem cell transplantation.
It is seen as enabling one-step-earlier intervention than the conventional method, which could assess risk only on or after the day of transplantation.
Seoul National University Hospital said on the 11th that a joint research team led by Pediatrics Professors Hong Gyeong-taek and Kang Hyeong-jin, and Department of Convergence Medicine Professor Han Do-hyun, with former researcher Yu Su-wan, developed an AI prediction model that can identify, at an early stage, high-risk groups for hepatic veno-occlusive disease (VOD) in pediatric hematopoietic stem cell transplantation patients. In the process, they also discovered key blood protein biomarkers that can predict disease onset.
Hematopoietic stem cell transplantation is a treatment that removes the damaged bone marrow of patients with blood disorders such as leukemia and transplants healthy hematopoietic stem cells to generate new blood cells. In pediatric patients, high-intensity anticancer conditioning is given before the transplant to clear diseased bone marrow, and highly toxic anticancer drugs used in this process, such as Busulfan, can damage hepatic microvasculature and cause serious complications.
A representative complication is hepatic veno-occlusive disease. It causes hepatomegaly, ascites, thrombocytopenia, and impaired liver and kidney function, and is known to occur in about 15% to 30% of pediatric patients undergoing hematopoietic stem cell transplantation. In particular, if it progresses to severe disease, it is fatal, with mortality reaching up to 80%.
To proactively administer preventive treatment by screening high-risk patients in advance, the team set out to discover blood protein biomarkers that can identify at-risk patients at the pre-chemotherapy stage.
To that end, they conducted a detailed analysis of 720 blood proteins before and after chemotherapy in 51 pediatric patients who received high-intensity conditioning with Busulfan ahead of allogeneic hematopoietic stem cell transplantation using a haploidentical donor (a donor such as a parent who is a half match). The study subjects consisted of 26 patients who developed severe hepatic veno-occlusive disease and a control group of 25 who did not.
The analysis found that patients without complications already had higher levels, before anticancer treatment, of GCLC, an enzyme that detoxifies toxins in the liver. In other words, they had sufficient "detoxification capacity" to clear highly toxic anticancer drugs.
By contrast, despite receiving the same treatment, patients who developed the disease already had lower levels of that enzyme before starting anticancer treatment, and expression of FBP1, a protein reflecting hepatic metabolic function, was also significantly lower, indicating vulnerability to toxic stress.
Based on these results, the research team selected 15 early protein biomarkers capable of predicting disease onset and built a Machine Learning model. To enhance clinical applicability, they then reduced the panel to the five proteins with the highest predictive power (HRNR, FBP1, DCD, GCLC, LSAMP), and analysis showed that these markers alone delivered high predictive performance for distinguishing high-risk groups (AUC 0.922). An AUC closer to 1 indicates higher predictive power.
Hong Gyeong-taek, a pediatrics professor at Seoul National University Hospital, said, "We confirmed that patients who develop hepatic veno-occlusive disease already show a different blood proteome pattern before anticancer treatment," adding, "The proteomic markers identified in this study will be an important turning point for early screening of high-risk patients to implement preventive therapy and to establish safer transplant treatment strategies."
The findings were published online in the latest edition of Transplantation and Cellular Therapy, the official journal of the American Society for Transplantation and Cellular Therapy (ASTCT).
References
Transplantation and Cellular Therapy (2025), DOI: https://doi.org/10.1016/j.jtct.2025.12.989