A new study shows that analyzing cell-free DNA (cfDNA) fragment patterns in blood can detect liver disease earlier and predict patient survival. The method uses a liquid biopsy to identify tissue-specific damage without invasive procedures.
Researchers analyzed whole-genome cfDNA fragmentomes from 1,576 participants, including those with fibrosis, cirrhosis, and other conditions. A machine learning classifier successfully identified liver disease stages with high sensitivity and minimal cross-reactivity.
The model was trained on 423 patients and validated in 221 others, confirming reproducibility. Additional data combining fragmentome and methylome profiles revealed signals from both liver and immune cells, suggesting dual biological insights.
A second algorithm predicted overall survival using cfDNA patterns, tested in cohorts of 571 and 231 individuals. The findings support cfDNA fragmentomes as powerful biomarkers of physiological health.
Scientists warn that unbalanced training data may introduce sex-based biases in machine learning models. Access to sex-disaggregated data is critical to ensure equitable performance across populations.