Reliable detection of Host-Microbe Signatures in cancer using PRISM

Recent controversy in the cancer microbiome field highlights the need for more reliable microbial detection from human genomic data. Here, we develop PRISM, an efficient computational framework for precise microorganism identification and decontamination from low-biomass sequencing data. PRISM achieves robust performance when benchmarked on 230 independent datasets with known true-positive and contaminant taxa. We then use PRISM to profile 25 cancer types from The Cancer Genome Atlas and Clinical Proteomic Tumor Analysis Consortium. We identify consistent microbial signatures in gastrointestinal tract, head-and-neck, and urogenital tract tumors, and sparse signal elsewhere. In pancreatic cancer, we associate microbial detection with altered host protein glycosylation pathways and greater smoking exposure. Lastly, we consider the impact of sequencing approaches on positive and negative data interpretation. Overall, PRISM improves the reliability of microbial profiling and allows leveraging of existing human genomic data for the concurrent detection of host-microbial signatures with potential molecular and clinical significance.