Cancer
Pooled analysis of cell-free DNA derived 5-hydroxymethylcytosine signatures and risk of multiple myeloma using Bayesian machine learning Saurabh Bhandari* Saurabh Bhandari Bhandari Bhandari University of Chicago
Background: Epigenetic changes play a critical role in the development of multiple myeloma (MM), but most studies use surrogate tissues (e.g., peripheral blood lymphocytes because obtaining tumor cells from the bone marrow in controls is not feasible) and do not distinguish 5-hydroxymethylcytosine (5hmC), a distinct epigenetic mark involved in gene activation, from the common-studied 5-methylcytosine due to technical constraints.
Methods: To elucidate novel 5hmC contributions to the risk of MM, we conducted a pooled case-control analysis of cell-free DNA (cfDNA) derived genome-wide 5hmC profiles from 575 MM cases and 294 controls across two independent studies. We applied semi-parametric Bayesian Additive Regression Trees integrating high-dimensional genomic features with confounders, using Bayesian false discovery rate control for variable selection. Model performance was evaluated in a held-out validation set (n=369) using area under the receiver operating characteristic curve (AUC) and Brier score.
Results: We identified eight genes (IL1RAP, ST5, HERC6, KL, MYO1E, ELK3, CAPN2, UBR4) associated with risk of MM, all with posterior inclusion probability of 1.00. The model achieved excellent discrimination in validation (AUC: 0.96; 95% credible interval: 0.95-0.97) with good calibration (Brier score: 0.085). Seven genes were independently confirmed by Elastic Net regularization. The identified genes implicate established MM pathogenic pathways, including interleukin-1 inflammatory signaling, fibroblast growth factor 23/Klotho-mediated bone destruction, and ubiquitin-proteasome dysregulation.
Conclusions: Using a Bayesian machine learning approach, we identified eight 5hmC-modified genes in cfDNA that are associated with risk of MM. These findings not only improve our understanding of epigenetic mechanisms that underlie MM risk but may also provide novel approaches for individualized prevention of this incurable malignancy.