UMI-tools: Modelling sequencing errors in Unique Molecular Identifiers to improve quantification accuracy

Abstract

Unique Molecular Identifiers (UMIs) are random oligonucleotide barcodes that are increasingly used in high-throughout sequencing experiments. Through a UMI, identical copies arising from distinct molecules can be distinguished from those arising through PCR amplification of the same molecule. However, bioinformatic methods to leverage the information from UMIs have yet to be formalised. In particular, sequencing errors in the UMI sequence are often ignored, or else resolved in an ad-hoc manner. We show that errors in the UMI sequence are common and introduce network-based methods to account for these errors when identifying PCR duplicates. Using these methods, we demonstrate improved quantification accuracy both under simulated conditions and real iCLIP and single cell RNA-Seq datasets. Reproducibility between iCLIP replicates and single cell RNA-Seq clustering are both improved using our proposed network-based method, demonstrating the value of properly accounting for errors in UMIs. These methods are implemented in the open source UMI-tools software package (https://github.com/CGATOxford/UMI-tools).

Batch effects and the effective design of single-cell gene expression studies

“…First, we combined all reads for a given single cell using samtools42. Next, we converted read counts to molecule counts using UMI-tools43. UMI-tools counts the number of UMIs at each read start position.…”

Gene-expression profiling of single cells from archival tissue with laser-capture microdissection and Smart-3SEQ

“…The incorporation of UMIs before PCR allows distinguishing coincidental fragmentation duplicates from true PCR duplicates among reads that align to the same position. However, as reported previously [37], the simplistic approach of allowing only one hit per UMI per position also fails when the read density is high and the number of possible UMIs is low, such that even duplicates with the same genome position and same UMI may occur by chance.…”

Using single‐cell genomics to understand developmental processes and cell fate decisions

“…Notably, barcode decomposition is not trivial—particularly for the random sequences of UMIs—as sequencing errors can alter their observed sequences. Methods have been developed to account for this by predicting which barcodes have arisen by error and which truly existed within the sample (Smith et al , ).…”