Noninvasive Diagnosis of Cancer Based on the Heterogeneity and Fragmentation Features of Cell-Free Mitochondrial DNA

Circulating cell-free DNA (cfDNA) has been widely studied in liquid biopsy for cancer detection and monitoring. Studies have shown that the mitochondrial DNA from tumor cells can be detected in plasma cfDNAs, and cell-free mitochondrial DNA (mtDNA) may become potential markers to screen cancer. Based on the mitochondrial heterogeneity, we design three types of heterogeneous features for mtDNA, including mtDNA copy number mutation, trinucleotide mutations, and point heteroplasm. Further, two types of fragmentation features of mtDNA, end motif and breakpoint motif, are also applied to cancer detection. The results show that the three types of heterogeneity features of mtDNA have better performance than the fragmentation features in cancer detection. The features of cell-free mtDNA designed in this study are superior than the mtDNA score proposed in the existing studies. Moreover, a stacking ensemble model is built to integrate the heterogeneity features and fragmentation features of cell-free mtDNA, and has good classification performance in cancer detection with accuracy of 0.9, specificity of 0.93, recall of 0.83, \(F1-score\) of 0.83, and AUC value of 0.98. The heterogeneity and fragmentation features of cell-free mtDNA and the cancer detection model constructed in this study provide a new angle for the liquid biopsy based cancer detection and early screening.