Table 3. Time taken to deduplicate a BAM file with many UMIs spread out over different alignment coordinates.

The n-grams BK-trees data structure is used in UMICollapse. Run time is measured in seconds. Default settings (i.e., directional algorithm, allowing one edit) are used for both UMI-tools and UMICollapse. The best time for each dataset is bolded.

Dataset	UMIs	UMI-tools	UMICollapse
Kivioja et al. (2012)	1,338,610	41.46	8.732
Müller-McNicoll et al. (2016)	1,175,027	8.69	4.26
Müller-McNicoll et al. (2016)	12,925,297	94.13	34.43
Müller-McNicoll et al. (2016)	118,677,727	889.42	271.31