Harnessing Metabolites as Serum Biomarkers for Liver Graft Pathology Prediction Using Machine Learning.
Cristina Baciu1, Soumita Ghosh1, Sara Naimimohasses1
1Ajmera Transplant Program, University Health Network, Toronto, ON M5G 2C4, Canada.
Metabolites
|May 24, 2024
Summary
A new machine learning model integrating serum metabolites and clinical data shows promise for non-invasively diagnosing liver transplant graft injury. This biomarker tool can help identify metabolic dysfunction-associated steatohepatitis, T-cell mediated rejection, and biliary complications.
Area of Science:
- Hepatology
- Transplant Immunology
- Biomarker Discovery
Background:
- Liver transplant (LT) recipients frequently experience graft injury, impacting over 50% of patients.
- Current diagnostic methods for graft injury lack non-invasive biomarkers, necessitating improved approaches.
- Distinguishing between metabolic dysfunction-associated steatohepatitis (MASH), T-cell mediated rejection (TCMR), and biliary complications is clinically significant.
Purpose of the Study:
- To develop a non-invasive serum biomarker for diagnosing liver transplant graft injury.
- To integrate serum metabolomic profiles with clinical variables using machine learning.
- To differentiate between MASH, TCMR, and biliary complications in LT recipients.
Main Methods:
- Serum samples from 55 LT recipients with biopsy-confirmed MASH, TCMR, or biliary complications were analyzed.
- Liquid chromatography-tandem mass spectrometry (LC-MS/MS) was used for metabolomic profiling.
- A multi-class machine learning classifier integrated metabolomic data and clinical variables.
Main Results:
- The machine learning model achieved an overall accuracy of 79.66% in classifying graft pathologies.
- The model demonstrated high accuracy in identifying MASH (7.4% OOB error), with lower accuracy for biliary (22.2%) and TCMR (29.6%) complications.
- Serine and serotonin were identified as key predictive metabolites, with significant Area Under the Curve (AUC) values for binary outcome predictions (0.882-0.972).
Conclusions:
- An integrated machine learning tool utilizing serum metabolites and clinical data shows potential as a non-invasive, multi-class biomarker for liver transplant graft pathology.
- This approach offers a promising avenue for improved diagnosis and management of graft injury in liver transplant recipients.
- Further validation is warranted to establish this tool in clinical practice for liver transplant graft monitoring.
