Background/objectives:
Early detection can significantly improve clinical outcomes for a number of cancers, but many of the best current screening methods require invasive procedures. A promising alternative approach is a liquid biopsy of cell-free DNA (cfDNA) from plasma. Because tumors generally shed relatively large amounts of DNA into the circulation, cancer can potentially be detected by identifying oncogenic variants in cfDNA. This process generally requires extremely deep sequencing, and in many cases is limited by the accuracy of next-generation sequencing (NGS). 

Methods:
One approach to overcoming this limitation is the use of unique molecular identifiers (UMIs), short sequences that uniquely tag each input DNA molecule prior to preparing NGS libraries. The approach can further be improved by tagging each original strand of the DNA molecule, in a technique termed duplex sequencing, which can correct early PCR errors and/or single-strand DNA damage events. Here we describe a new library preparation system incorporating short, discrete UMI sequences to maximize sequence distances for error correction. 

Results:
We show that this system can determine the conversion efficiency of NGS libraries. Using the Twist cfDNA Pan-cancer Reference Standards to simulate a low fraction of tumor DNA in a healthy background, we demonstrate high sensitivity towards a variety of oncogenic substitutions, indels and structural variants. We demonstrate the baseline error rate using unmodified human cfDNA, and use the system to determine the mutation frequency in a synthetic biology application. 

Conclusion:
In summary, this study demonstrates the utility of UMIs for a variety of applications in NGS.