Circulating Tumor DNA Methylation with Photoelectrochemical Sensor Signatures Decoded by Attention Mechanisms

Integration of ultrasensitive photoelectrochemical (PEC) sensing with advanced computational analysis presents a novel approach for early cancer detection through circulating tumor DNA (ctDNA) methylation profiling. We report a synergistic platform combining BiOBr/WS2 heterojunction–based PEC sensor with attention–driven deep learning to decode cancer–specific methylation signatures. The BiOBr/WS2 composite, synthesized via a two–step hydrothermal process, demonstrated enhanced visible–light absorption and charge separation efficiency, achieving a 2.3–fold higher photocurrent response than pristine BiOBr. This sensor enabled ultrasensitive ctDNA detection with a linear range spanning six orders of magnitude (0.5 fM to 500 pM) and an ultralow detection limit of 0.05 fM. Deep learning revealed that attention mechanisms prioritized methylation loci in promoter regions and gene bodies, achieving 75% classification accuracy and a ROC AUC of 0.889. The model identified bidirectional methylation patterns, with 73% of high–weight CpG sites showing cancer–specific hypomethylation, challenging traditional hypermethylation–centric paradigms. The sensor exhibited exceptional selectivity against mismatched sequences, 95% signal stability over 7 days, and 97–103% recovery in human serum. By bridging nanomaterial–enhanced PEC transduction with context–aware methylation analysis, this work establishes robust framework for non–invasive cancer screening, emphasizing the critical role of epigenetic heterogeneity in early intervention strategies.