Advances in liquid biopsy for preeclampsia risk prediction and tumor gene expression inference from cell-free DNA

Abstract

Liquid biopsy has the potential to address critical diagnostic needs; however, substantial innovations are still required. Here we describe two unique approaches that explore epigenetic features of cell-free DNA using practical, scalable, and cost-effective assays to infer clinically relevant insights. In the first study, we leveraged a routine prenatal cfDNA screening (PDNAS) assay to predict the risk for preeclampsia early in pregnancy up to 5 months before onset of clinical symptoms with an area under the receiver operating characteristic curve (AUC) 0.85. For this we developed the PEARL (preeclampsia early assessment of risk from liquid biopsy) framework which infers nucleosome accessibility to determine placental and endothelial dysfunction in a non-invasive manner. In the second study, we expand the diagnostic spectrum of liquid biopsies by developing DRAGN-X (Deep Regulatory Autoencoder for Gene Expression), a framework that infers quantitative, tumor-specific gene expression directly from cfDNA methylation. DRAGN-X integrates novel insights into the relationship between DNA methylation and gene expression with deep learning models, achieving a high correlation (Pearson’s r = 0.91) between cfDNA-inferred expression and ground-truth tumor tissue expression. We demonstrate broad clinical utility of DRAGN-X across multiple cancer and liquid biopsy types using a scalable, cost-efficient, low-pass whole-genome cfDNA methylation assay. Together, these findings establish advances in cfDNA technologies that can enhance precision medicine and improve clinical diagnostics.