lh3 / unimap

Getting Started

# compile and install
git clone https://github.com/lh3/unimap
cd unimap && make

# simple use cases (-b24 below saves memory for this toy example)
./unimap -b24 -c test/MT-human.fa test/MT-orang.fa > out.paf

# prebuild index; recommended as unimap indexing is slower than minimap2 indexing
./unimap -b24 -d MT-human.umi test/MT-human.fa
./unimap -c MT-human.umi test/MT-orang.fa > out.paf

# use presets (no test data)
unimap -cxasm5 --cs -t8 ref.fa asm.fa         # if asm.fa is near identical to ref.fa
unimap -cxhifi --cs -t8 ref.fa hifi-reads.fa  # HiFi reads to reference alignment
unimap -cxont  --cs -t8 ref.fa ont-reads.fa   # Nanopore reads to reference alignment

Introduction

Unimap is a fork of minimap2 optimized for assembly-to-reference alignment. It integrates the minigraph chaining algorithm and can align through long INDELs (up to 100kb by default) much faster than minimap2. Unimap is a better fit for resolving segmental duplications and is recommended over minimap2 for alignment between high-quality assemblies.

Unimap does not replace minimap2 for other types of alignment. It drops the support of multi-part index and short-read mapping. Its long-read alignment is different from minimap2 but is not necessarily better. Unimap is more of a specialized minimap2 at the moment.

Notes

• With the default asm5 preset, unimap may align a highly diverged region as a long insertions followed by a long deletion. Truvari may identify two false positive calls in this case, but these arguably are not errors.

• Unimap takes ~5 minutes to index a human genome, slower than minimap2. It is recommended to save the index for faster startup.

• The default ont preset has been tuned for more recent Nanopore reads at ~95% accuracy.