Cell-free DNA (cfDNA) is a promising non-invasive biomarker for cancer detection; however, its clinical utility is often limited by its low abundance, short and heterogeneous fragment lengths. Here, we present a label and amplification-free strategy for cfDNA analysis and colorectal cancer (CRC) stage classification using an ultra-small (<3 nm) solid-state nanopore probed by high-bandwidth electronics. We analyzed cfDNA samples from healthy donors and CRC patients across multiple independent experimental runs to assess reproducibility. While conventional event-averaged descriptors captured broad group-level differences, they were insufficient for reliable cancer staging. In contrast, analysis of the high-temporal-resolution ionic current fluctuations revealed heterogeneous fragment populations and molecular features preserved in the native cfDNA. To decode these signatures, we developed a hybrid convolutional neural network (CNN)-transformer framework that directly processes raw current time series. This approach achieved ~95% accuracy in distinguishing healthy from CRC samples and enabled discrimination among cancer stages, significantly outperforming feature-averaged metrics and length-based baselines. Blind testing on previously unseen samples yielded consistent healthy vs. sick and stage-specific predictions, demonstrating robust generalization across patient cohorts. Together, these results establish high-bandwidth nanopore sensing coupled with time-resolved machine learning as a potential point of care framework for direct cfDNA profiling.
Pal et al. (Wed,) studied this question.