---
title: Decoding UMI treatment and metrics
description: Understand UMI processing, and how counting final UMI 'groups' versus initial 'reads' explains the difference between singleton percentages and duplication rates.
---
[Skip to content](https://hubspot.seqone.com/test_seqone/decoding-umi-treatment-and-metrics#main-content)
English
Show submenu for translations
SeqOne
Open main navigation
Close main navigation
- English
Show submenu for translations
- Contact us
Contact us
How can we help you?
- There are no suggestions because the search field is empty.
1. [Test\_SeqOne](https://hubspot.seqone.com/test_seqone?hsLang=en)
2. [Analysis & Interpretation](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en)
3. [Coverage and Metrics](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#coverage-and-metrics)
# Decoding UMI treatment and metrics
## Understand UMI processing, and how counting final UMI 'groups' versus initial 'reads' explains the difference between singleton percentages and duplication rates.
In next-generation sequencing (NGS), a common challenge is distinguishing between a true biological signal and a technical artifact. The main source of these artifacts is the **PCR**, a process used during library prep to amplify small amounts of DNA. PCR creates multiple copies of the same DNA fragment, known as **PCR duplicates**.
Traditional bioinformatic tools handle this by identifying and removing reads that align to the exact same start and end position in the genome. This "naïve" deduplication approach is simplistic and can mistakenly discard reads from different original molecules that just happen to align identically. This becomes a major problem when trying to detect true, low-frequency variants.
---
### The UMI Solution: Tagging, Grouping, and Consensus
UMI-based methodologies provide a more sophisticated solution by uniquely tagging each DNA molecule *before* amplification. This allows for precise tracking of each original fragment.
Here’s how it works:
#### 1. Tagging Before Amplification
A Unique Molecular Identifier (UMI) is a short DNA sequence, or tag, that is ligated to each DNA fragment during the library preparation phase *before PCR*. This crucial step ensures that every individual molecule from the original sample receives its own unique barcode before it gets amplified.
#### 2. Grouping by Family
After sequencing, all the resulting reads are analyzed. Reads that map at the same spot on the genome *and* share the same UMI sequence, are recognized as having originated from the same single parent molecule and are grouped into a "read family".
#### 3. Building a Consensus
Instead of just discarding all but one read in a family, the UMI approach merges all reads within the same group to create a single, high-fidelity **consensus sequence**. This process works base-by-base, using a probability model that accounts for the quality score of each base in every read to determine the correct sequence.
This consensus approach provides two major advantages:
- **Accurate Deduplication**: It correctly identifies and collapses reads that are true PCR duplicates, reducing the final data to the correct number of original molecules.
- **Error Correction**: By comparing all the reads in a family, random errors introduced during PCR or sequencing can be identified and filtered out. A true variant present in the original molecule will appear in all the reads of that UMI family, while a random error will only appear in one or a few, allowing it to be dismissed.
In the SeqOne platform, reads with the same UMI are used to generate a consensus when there are two or more reads in a group. Reads from UMIs that were sequenced only once (singletons) are also kept. The final, processed dataset is composed of a mix of the singletons and the higher-confidence "groups", which provides increased confidence in low-frequency variant calls and results in higher-quality data overall.
---
### Decoding UMI Metrics: Reads vs. Groups
When analyzing sequencing data that uses UMI, it's common to see a significant drop in the total number of reads after processing. At the same time, you might notice a high percentage of "singletons" in your results. This can be confusing. If, for instance, 79% of your final data consists of singletons (groups made of a single read), why isn't the read duplication rate simply 21% (100% - 79%)?
The answer lies in understanding what each metric is actually counting. They are two distinct, yet complementary, quality control metrics.
- **Read Duplication (% Drop in Reads):** This number reflects the total volume of raw sequencing reads that were identified as PCR duplicates and merged into consensus groups. It is a measure of library complexity and amplification efficiency, calculated based on the *initial* number of reads.
- **Singleton Percentage (% Singletons):** This number describes the composition of the *final*, deduplicated dataset. It tells you what proportion of your final "read groups" are made up of just a single, unique read (as opposed to groups that were built from the consensus of several duplicated reads).
A group can be either a **singleton** (composed of one read) or a **consensus** (composed of two or more reads that have been merged). The duplication percentage counts the *reads* that were collapsed, while the singleton percentage counts the resulting *groups*.
#### Example
For example, the library plotted here has ~3 Million reads and 48.4% duplication: this means that the UMI deduplication steps removes ~1.5 million reads from the *raw library*, by merging them in consensus of 2, 3, 4 or multiple reads.
The barplot is a visualization after merging: 78.89% of the *final reads* are singletons, which means they were not duplicated to begin with. The higher-quality consensus reads (with info collapsed from several duplicates, so a *reduced* number of reads compared to the raw counts) make up 21% of the final library.
- [Releases](https://hubspot.seqone.com/test_seqone/releases?hsLang=en#main-content)
- [Spring 2026](https://hubspot.seqone.com/test_seqone/releases?hsLang=en#spring-2026)
- [Release Notes 2025](https://hubspot.seqone.com/test_seqone/releases?hsLang=en#release-notes-2025)
- [What's new in Preview?](https://hubspot.seqone.com/test_seqone/releases?hsLang=en#whats-new-in-preview)
- [Analysis & Interpretation](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#main-content)
- [Variant table](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#variant-table)
- [Workflow Management](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#workflow-management)
- [Annotation and Scores](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#annotation-and-scores)
- [Reporting](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#reporting)
- [DiagAI](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#diagai)
- [Coverage and Metrics](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#coverage-and-metrics)
- [CNV](https://hubspot.seqone.com/test_seqone/analysis-interpretation?hsLang=en#cnv)
- [Getting started](https://hubspot.seqone.com/test_seqone/getting-started?hsLang=en#main-content)
- [General Platform Usage](https://hubspot.seqone.com/test_seqone/getting-started?hsLang=en#general-platform-usage)
- [Intro to worksets](https://hubspot.seqone.com/test_seqone/getting-started?hsLang=en#intro-to-worksets)
- [Video Tutorials](https://hubspot.seqone.com/test_seqone/getting-started?hsLang=en#video-tutorials)
- [Files](https://hubspot.seqone.com/test_seqone/getting-started?hsLang=en#files)
- [Intro to Platform Features](https://hubspot.seqone.com/test_seqone/getting-started?hsLang=en#intro-to-platform-features)
- [Troubleshooting](https://hubspot.seqone.com/test_seqone/troubleshooting?hsLang=en#main-content)
- [General](https://hubspot.seqone.com/test_seqone/troubleshooting?hsLang=en#general)
- [Credentials](https://hubspot.seqone.com/test_seqone/troubleshooting?hsLang=en#credentials)
- [Error and info messages](https://hubspot.seqone.com/test_seqone/troubleshooting?hsLang=en#error-and-info-messages)
- [Information Security & Compliance](https://hubspot.seqone.com/test_seqone/information-security-compliance?hsLang=en#main-content)
- [General](https://hubspot.seqone.com/test_seqone/information-security-compliance?hsLang=en#general)
- [AI Systems Documentation](https://hubspot.seqone.com/test_seqone/ai-systems-documentation?hsLang=en#main-content)
- [DiagAI Autofill](https://hubspot.seqone.com/test_seqone/ai-systems-documentation?hsLang=en#diagai-autofill)
- [DiagAI HPO](https://hubspot.seqone.com/test_seqone/ai-systems-documentation?hsLang=en#diagai-hpo)
- [Technical Information](https://hubspot.seqone.com/test_seqone/technical-information?hsLang=en#main-content)
- [Workflows - Somatic](https://hubspot.seqone.com/test_seqone/technical-information?hsLang=en#workflows-somatic)
- [Technical details](https://hubspot.seqone.com/test_seqone/technical-information?hsLang=en#technical-details)
- [Databases](https://hubspot.seqone.com/test_seqone/technical-information?hsLang=en#databases)
# SeqOne
[mailto:email@email.com](mailto:email@email.com)
Copyright © 2026, SeqOne