UCLA

With the noninvasiveness of sample collection and the comprehensiveness of the DNA profile from various tissues, plasma cell-free DNA (cfDNA) has attracted enormous attention for many applications, including disease-related marker identification, disease detection, and disease monitoring. However, since cfDNA is a mixture of disease-related DNA in an overwhelming pool of DNA from normal cells, the weak disease signal poses a major challenge for these applications. Current methods usually employ traditional error suppression for genomic DNA samples and deep sequencing on small panels, which limit their performance. A fundamental and yet underdeveloped task for these applications is the precise and sensitive calling of somatic single nucleotide variants (SNVs) from cfDNA. We present cfSNV, a somatic SNV detection method designed specifically for cfDNA that incorporates multilayer error suppression and hierarchical mutation calling. The accurate and sensitive identification of disease-related markers can provide a reliable foundation for disease monitoring, which is essential for assessing the effectiveness of treatment. We provide a novel cancer monitoring approach, OncoMonitor, which comprehensively analyzes tumor mutations and sensitively detects minimal residual disease, cancer recurrence, second primary diseases, and cancer progression with longitudinal cfDNA samples. Further leveraging the information in cfDNA samples, we developed a workflow using the microbiome composition in cfDNA for disease detection, which provides complementary disease evidence to current human-origin cfDNA-based methods. In summary, this work uses statistical methods and machine learning models to address the current limitations in mutation detection and disease monitoring in cfDNA and provide complementary information for disease detection.