Integrative multi-omics data fusion for potential biomarker prioritization in sulfur mustard exposed patients: A comparative study

Background Sulfur mustard exposure causes severe and heterogeneous long-term health complications, yet the molecular determinants of disease severity remain incompletely understood. Integrative multi-omics approaches provide an opportunity to characterize coordinated biological responses and identify prioritized candidate biomarkers severity-associated biomarkers. Methods Whole-blood transcriptomic, genome-wide DNA methylation, and serum metabolomic profiles were generated from 52 individuals stratified by clinical severity (healthy controls and mild, moderate, and severe injury groups). Five complementary integration frameworks (sPLS, NMF, MOFA+, SNF, and JIVE) were applied to a biologically curated feature set. Candidate biomarkers were prioritized using the complementary outputs of sPLS, NMF, and MOFA+, followed by univariate differential analysis. Consensus biomarkers were subsequently identified by integrating differential analysis, SNF-derived, and JIVE-derived evidence, retaining features supported by at least two evidence sources. Random Forest was then used for post hoc biomarker prioritization. Results Among the evaluated methods, supervised sPLS showed the highest agreement with clinical severity classifications (Adjusted Rand Index = 0.441), whereas NMF and JIVE identified biologically coherent patient subgroups. JIVE decomposition revealed that 57.24% of total multi-omics variance was shared across transcriptomic, epigenomic, and metabolomic layers, indicating predominantly coordinated molecular dysregulation. Combined differential analysis across sPLS, NMF, and MOFA+ identified 589 candidate biomarkers. Consensus analysis incorporating evidence from all five integration frameworks yielded a panel of 98 candidate biomarkers supported by at least two independent methods. Functional enrichment highlighted signatures related to immune activation, receptor signaling modulation, neuro-sensory pathways, tissue remodeling, and epigenetic reprogramming. Notably, CCNJL was the only biomarker consistently retained across all five integration frameworks. Random Forest prioritization further highlighted LOC112268313, LOC102724621, and multiple rRNA transcripts among the most informative severity-associated features. Conclusions Multi-omics integration revealed a predominantly shared molecular architecture underlying sulfur mustard-induced injury. The resulting 98-feature consensus signature provides a biologically informed set of candidate biomarkers for future validation studies and mechanistic investigations of sulfur mustard injury severity.