Integration of breaking-point transition probabilities and specific open chromatin region features for early cancer detection.

cfDNA fragments exhibited strong sequence bias within 16 bp around breakpoints, extending beyond the conventional 5'-end motif window.

Using features derived from two external datasets (n = 129 and n = 225), the SOTP_BP16_OCR model achieved AUCs of 0.94 and 0.99 for case-control classification.

Tissue-of-origin prediction, which could only be performed in one dataset, reached an accuracy of 0.802.