fgumi

⚠️ RESEARCH PREVIEW - USE AT YOUR OWN RISK

This software is currently a research preview. While we have extensively tested these tools across a wide variety of vendor-provided data, no guarantees are made regarding correctness or stability.

We are targeting July 31, 2026 to recommend fgumi over fgbio for production use.

Fulcrum Genomics Unique Molecular Indexing (UMI) Tools - a suite of high-performance tools for working with UMI-tagged sequencing data.

Visit us at Fulcrum Genomics to learn more about how we can power your Bioinformatics with fgumi and beyond.

Overview

fgumi provides comprehensive functionality for:

• UMI extraction from FASTQ files
• Read grouping by UMI with multiple assignment strategies
• UMI-aware deduplication for marking/removing PCR duplicates
• Consensus calling (simplex, duplex, and CODEC)
• Quality filtering of consensus reads
• Read clipping for overlapping pairs
• Metrics collection for QC and analysis

Pipeline Overview

The diagram shows the workflow from FASTQ files to filtered consensus reads:

• Red: Simplex (single-strand) consensus
• Blue: Duplex (double-strand) consensus
• Green: CODEC consensus
• Orange: Optional UMI correction for fixed UMI sets

Resources

• Documentation
• Best Practice Pipeline: Recommended workflow from FASTQ to consensus
• Performance Tuning Guide: Threading, memory, and compression optimization
• Snakemake Pipeline: Reference implementation
• Metrics: Output metrics documentation
• Developing: Developer guide
• Compare CLI: Compare command documentation (feature-gated)
• Simulate CLI: Simulate command documentation (feature-gated)
• Releases
• Issues: Report a bug or request a feature
• Pull requests: Submit a patch or new feature
• Discussions: Ask a question
• Contributors guide
• License: Released under the MIT license

Installation

Downloading a pre-built binary

Pre-built binaries for the most common operating systems and CPU architectures are attached to each release for this project.

Installing with Cargo

cargo install fgumi

Building from source

Clone the repository:

git clone https://github.com/fulcrumgenomics/fgumi

Build the release version:

cd fgumi
cargo build --release

Optional Features

Feature	Description
compare	Developer tools for comparing BAMs and metrics
simulate	Commands for generating synthetic test data
profile-adjacency	Enable profiling output for adjacency UMI assigner
Enable with: cargo build --release --features <feature>

Available Tools

Command	Description	Equivalent Tool(s)
extract	Extract UMIs from FASTQ files	fgbio ExtractUmisFromBam
correct	Correct UMIs based on sequence similarity	fgbio CorrectUmis
zipper	Restore original FASTQ from unaligned BAM	fgbio ZipperBams, picard MergeBamAlignment
fastq	Convert BAM to FASTQ format	samtools fastq
sort	Sort BAM by coordinate/queryname/template	—
group	Group reads by UMI	fgbio GroupReadsByUmi
dedup	Mark/remove UMI-aware duplicates	gatk UmiAwareMarkDuplicatesWithMateCigar, umi-tools dedup
simplex	Call single-strand consensus reads	fgbio CallMolecularConsensusReads
duplex	Call duplex consensus reads	fgbio CallDuplexConsensusReads
codec	Call CODEC consensus	fgbio CallCodecConsensusReads
filter	Filter consensus reads	fgbio FilterConsensusReads
clip	Clip overlapping read pairs	fgbio ClipBam
duplex-metrics	Collect duplex metrics	fgbio CollectDuplexSeqMetrics
review	Review consensus variants	fgbio ReviewConsensusVariants
downsample	Downsample BAM by UMI family	N/A
compare <cmd>	Compare files (feature-gated)	N/A
simulate <cmd>	Generate test data (feature-gated)	N/A

Usage

For detailed usage of each command, run:

fgumi <command> --help

Basic Workflow

1. Extract UMIs from FASTQ:

fgumi extract \
  --inputs R1.fastq.gz R2.fastq.gz \
  --read-structures +T +M \
  --output unaligned.bam \
  --sample MySample \
  --library MyLibrary

2. (Optional) Correct UMIs for fixed UMI sets:

fgumi correct \
  --input unaligned.bam \
  --output corrected.bam \
  --umi-files umis.txt \
  --min-distance 1

3. Align and sort reads using fgumi fastq + zipper + sort pipeline:

fgumi fastq --input unaligned.bam \
  | bwa mem -p ref.fa - \
  | fgumi zipper --unmapped unaligned.bam \
  | fgumi sort --output sorted.bam --order template-coordinate

4. Group reads by UMI:

fgumi group \
  --input sorted.bam \
  --output grouped.bam \
  --strategy paired   # for duplex workflows
  # or --strategy adjacency for simplex/codec workflows

5. Call consensus reads:

# Simplex consensus
fgumi simplex \
  --input grouped.bam \
  --output consensus.bam

# Or duplex consensus
fgumi duplex \
  --input grouped.bam \
  --output duplex.bam

# Or codec consensus
fgumi codec \
  --input grouped.bam \
  --output codec_consensus.bam

6. (Optional) Collect duplex metrics:

fgumi duplex-metrics \
  --input grouped.bam \
  --output metrics

7. Filter consensus reads:

fgumi filter \
  --input consensus.bam \
  --output filtered.bam \
  --ref ref.fa \
  --min-reads 1,1,1

Performance Options

fgumi supports multi-threading and memory management for optimal performance:

• Threading: --threads N for parallel processing
• Memory: --max-memory 768 (plain numbers are MiB; supports human-readable formats like 2GB; pass auto to size to the host)
• Compression: --compression-level 1-12 for speed vs size trade-offs

Memory Model — Important for fgbio users: Unlike fgbio's JVM -Xmx which sets a hard ceiling on total process memory, fgumi's --max-memory controls pipeline queue backpressure only. Two things to be aware of:

1. Per-thread scaling (default): --max-memory 768 --threads 8 allocates 768 MiB per thread = ~6 GB total queue memory. Use --memory-per-thread false for a fixed total budget, or --max-memory auto to size the budget to the host (cgroup-aware, minus --memory-reserve) so the command self-throttles instead of OOM-ing.
2. Queue memory < total process memory: Actual RSS will be higher due to UMI data structures, decompressors, thread stacks, and working buffers.

See the Performance Tuning Guide for detailed guidance, including scenario-based configurations and troubleshooting.

Performance

fgumi is written in Rust for maximum performance.

Command-Level Optimizations

Command	Key Optimizations
extract	Work-stealing thread pool, streaming I/O
correct	N-gram indexing with pigeonhole principle, BK-tree for k>1
group	2-bit UMI encoding, N-gram/BK-tree indexing, directed adjacency graph
simplex	Fast-path for unanimous consensus, parallel processing
duplex	Parallel duplex calling, efficient strand matching
codec	Parallel CODEC consensus
filter	Streaming filter with parallel processing
clip	Parallel overlap detection and clipping
sort	External merge sort, configurable memory limit

General Optimizations

• 2-bit DNA encoding: 4 bases in 1 byte, 32 bases in u64
• CPU intrinsics: XOR + popcount for Hamming distance
• Work-stealing scheduler: Unified pipeline with dynamic load balancing
• libdeflate: Fast BGZF compression

Acknowledgements

fgumi's UMI grouping algorithms are inspired by:

• UMI-tools (Smith et al. 2017) - The directed adjacency method for UMI deduplication with count gradient constraints.
• UMICollapse (Liu 2019) - N-gram and BK-tree indexing strategies for efficient similarity search in UMI deduplication.

Authors

• Nils Homer
• Tim Fennell

Sponsors

Development of fgumi is supported by Fulcrum Genomics.

Become a sponsor

Disclaimer

This software is under active development. While we make a best effort to test this software and to fix issues as they are reported, this software is provided as-is without any warranty (see the license for details). Please submit an issue, and better yet a pull request as well, if you discover a bug or identify a missing feature. Please contact Fulcrum Genomics if you are considering using this software or are interested in sponsoring its development.