PROJECT SUMMARY/ABSTRCT Though hepatocellular carcinoma (HCC) is the 14th most common cancer in the US,1 it is the fifth most common cause of cancer deaths with a 5-year survival of 19.6%. The incidence of HCC in the US is rising at the annual rate of 4.5% per year, and it is the only cancer with an increase in mortality over the last ten years. Currently, the only method for detection of HCC is with surveillance imaging in patients with cirrhosis and alpha fetoprotein (AFP). However, approximately 13-20% of patients diagnosed with HCC do not have cirrhosis, and hence were never screened for the disease. In addition, AFP is normal in approximately 30-40% of patients with HCC. Because of this, HCC is usually detected at an advanced stage, when there are a limited number of therapeutic options. This proposal is born out of convincing preliminary data that indicate that DNA from tumor- dwelling microbes can identify malignancies. Approximately 2.5% of reads in The Cancer Genome Atlas (TCGA) are microbial. Machine learning algorithms using these microbial reads accurately identified solid tumors from each other and from adjacent control tissue. Moreover, much of this accuracy is maintained when blood samples are used instead of the tumors. This preliminary data is particularly robust for HCC. Based on this preliminary data, the proposed studies will advance the development of a biomarker for early detection of HCC in several ways. First, we will use the machine learning algorithms developed from TCGA on a new cohort of samples from the University of Florida liver biobank. This biobank has nearly as many HCC samples as the entire TCGA network. Second, the proposed studies will use more rigorous controls than what was available in the TCGA network. This includes the use of non-HCC liver malignancies and benign tumors. Whereas the machine learning algorithms were adequate in distinguishing HCC masses from other primary tumors in TCGA, it's not clear whether they can distinguish HCCs from other malignant and benign masses in the liver (e.g., metastatic colorectal cancer, hepatomas), which are more common than HCC, or the blood from patients with HCC to those from patients with cirrhosis without HCC. Finally, the proposed studies will help determine whether the machine learning algorithms developed from TCGA can detect whether HCC has been treated (e.g., chemoembolization) and thus potentially serve as a tool for surveillance. Overall, these experiments will help determine how generalizable the algorithms developed with TCGA are to independent cohorts of HCC. These studies will lay the foundation for the development of more effective screening and surveillance protocols that will hopefully impact the significant morbidity and mortality associated with HCC. Finally, if these studies are successful, they would encourage the exploration of using microbial DNA as an early detection biomarker for other types of cancers, and the role of tumor-dwelling bacteri...